Download the AI & Machine Learning For Dummies PRO App: iOS - Android Our AI and Machine Learning For Dummies PRO App can help you Ace the following AI and Machine Learning certifications:
Emerging AI Innovations: Top Trends Shaping the Landscape in September 2023.
In the dynamic landscape of Artificial Intelligence, September 2023 stands as a testament to the relentless pace of innovation. Titans of the industry such as Google, OpenAI, and Meta are at the forefront, unveiling cutting-edge developments with each passing day. The remarkable feats achieved by models like GPT are now being complemented by advancements in Lifelong Learning Machine Systems (LLMS). As we delve deeper into the realms of Generative AI and Discriminative AI, we are ushered into an era where machines don’t just compute—they conceptualize. Moreover, the rise of Explainable AI emphasizes the industry’s dedication to clarity and understanding, ensuring that as we integrate these AI systems more deeply into our daily lives, we remain fully informed about their decision-making processes. Join us as we navigate through the pivotal AI trends shaping September 2023 and discover how these technological wonders are transforming the world we live in.
Looking for a no-code tool to easily create and publish content? With Notice, generate custom FAQs, blogs, and wikis tailored to your business with AI in a single click.
Create, manage, and translate – all in one place. Collaborate with your team, and publish content across platforms, including CMS, HTML, or hosted versions.
Plus, you can enjoy cookie-free analytics to gain insights about users and enhance SEO with Notice‘s smart blocks. Use code DIDYOUNOTICE30SPECIAL for a 30% discount on any subscription.
AI Monthly Rundown September 2023: The Future of LLMs in Search!
In this blog, we’ll cover the evolution of search and large language models, Amazon’s investment in Anthropic and generative AI updates, Google’s advancements in personalized route suggestions and language modeling, DeepMind’s AlphaMissense system for predicting genetic diseases, OpenAI’s DALL·E 3 and the addition of voice and image capabilities to ChatGPT, Getty Images’ Generative AI art tool, Vectara’s Boomerang LLM, important AI contributions from Google and Apple, and the surge in hiring for Generative AI talent.
LLMs, or Large Language Models, have undoubtedly made a significant impact on the way we search for information. However, it’s unlikely that they will completely replace traditional search engines. Instead, I believe that LLMs and search engines will evolve and blend together to offer users the best of both worlds.
Search engines have been around for over 29 years and have become incredibly proficient at providing fresh, relevant, and comprehensive results. This level of expertise is the result of years of industrial research. On the other hand, LLMs, with their phenomenal language understanding and generation capabilities, offer a new dimension to information retrieval. However, they sometimes lack verifiability, attributions, and relevancy.
In my view, we can expect the rise of Language Search Models (LSMs), where the strengths of both LLMs and traditional search engines will be combined. This hybrid approach will leverage the comprehensiveness and freshness of search engines while harnessing LLMs’ advanced language processing abilities. By incorporating the best of both technologies, we can enhance the search experience and deliver even more relevant and accurate results to users.
References:
[1] The Anatomy of a Large-Scale Hypertextual Web Search Engine, 1998, Google
[2] Attention Is All You Need, 2017, Google
Guess what? Amazon has just announced a whopping $4 billion investment in Anthropic! This investment is part of their larger plan to collaborate and create top-notch foundation models for the industry.
Now, let me fill you in on the details. Anthropic is known for their cutting-edge safety research and products. And with Amazon Web Services (AWS) on board, they’re going to take their game to a whole new level. AWS, being experts in running secure and reliable infrastructure, will make Anthropic’s AI accessible to a wider audience.
But that’s not all! AWS will become Anthropic’s go-to cloud provider for mission-critical workloads. So, any heavy-duty tasks will rely on AWS’s powerful capabilities. Plus, this collaboration will further boost Anthropic’s support for Amazon Bedrock.
With Amazon’s massive investment, Anthropic and AWS are primed to create some seriously impressive technological advancements. And as a result, we can expect safer and more controllable AI systems. Talk about a win-win situation!
Get ready for some exciting developments in the world of artificial intelligence, my friend. This collaboration might just be a game-changer.
Hey there! Guess what? Google and DeepMind have come up with something super cool for Google Maps! They’ve developed this amazing AI algorithm that makes your route suggestions even more personalized. Can you imagine that?
This algorithm is no joke, it has a whopping 360 million parameters! It’s able to take into account real driving data from Maps users to figure out what factors are important to you when you’re making decisions about your route. It considers things like travel time, tolls, road conditions, and even your personal preferences. It’s like having your own personal navigator right in your pocket!
Now, you might be curious how this algorithm actually works. Well, it uses something called Inverse Reinforcement Learning (IRL) to learn from your behavior. And it also uses another fancy technique called Receding Horizon Inverse Planning (RHIP) for both short- and long-distance travel.
And the results are impressive! Tests have shown that the accuracy of suggested routes for two-wheelers has improved by a whooping 16 to 24 percent with the implementation of RHIP. And you know what’s even cooler? The algorithm is constantly learning and evolving, so it’s only going to get better at predicting the routes you prefer over time.
So, next time you’re hitting the road, keep an eye out for those hyper-personalized route suggestions powered by Google’s awesome AI algorithm. Happy travels!
So, there’s this fascinating survey that dives deep into the world of LLM-based agents. It’s like a one-stop shop for understanding these agents and how they can be a force for good. From the philosophical roots of agents to their development in AI, this survey covers it all.
But wait, what exactly are LLM-based agents, you ask? Well, they’re the perfect foundation for AI agents. And this survey presents a conceptual framework that can be customized for various applications. Talk about flexibility!
But that’s not all. This survey takes us on a journey through the extensive applications of LLM-based agents. We explore single-agent scenarios, multi-agent scenarios, and even how these agents can collaborate with humans. It’s all about teamwork, right?
And here’s the exciting part: we get to delve into agent societies. We’re talking about understanding the behavior and personality of LLM-based agents, the social phenomena that emerge when they form societies, and the valuable insights they bring to our own human society. It’s like a sneak peek into a future where AI agents and humans coexist.
To top it all off, this survey discusses key topics and challenges within the field. So, if you’re curious about the potential and rise of LLM-based agents, this survey is your go-to resource. It even paints a vivid picture of a society where humans and AI agents live harmoniously together. Incredible, right?
There’s a groundbreaking development in the world of 3D printing, and it involves the power of artificial intelligence. MIT researchers have come up with a tool called Style2Fab, which gives designers the ability to personalize their 3D models like never before.
Traditionally, when adding custom design elements to a 3D model, there has always been a risk of compromising the functionality of the final object. But with Style2Fab, that’s no longer an issue. This innovative tool uses generative AI to ensure that any design changes made still maintain the object’s intended functionality.
So how does it work? Well, it’s incredibly user-friendly. Designers simply need to provide natural language prompts describing the desired design elements they want to add. No need for complex technical skills or software expertise. Once the prompts are given, the AI takes over and generates the necessary modifications to the 3D model.
Once the personalized design is ready, it’s time to bring it to life with a 3D printer. The user can simply fabricate the object and voila! You have a custom-made, functional 3D-printed object that perfectly matches your vision.
This technology opens up endless possibilities for customization and personalization in 3D printing. With Style2Fab, designers can unleash their creativity without any limitations. So get ready to transform your ideas into tangible objects with the power of AI and 3D printing.
Google has come out with an even better version of Bard, and it’s got some exciting new features. Let me tell you about them.
First up, we’ve got Bard Extensions in English. This means that Bard can now dig into all the Google tools you use every day, like Gmail, Docs, Drive, Maps, YouTube, and even Google Flights and hotels. So, if you’re searching for something and the information is spread across multiple apps and services, Bard will be able to find and show you exactly what you need.
Next, we’ve got a little something called “Google it.” Now, when you’re browsing and Bard gives you an answer, you can easily double-check its accuracy. Just click on the “G” icon and Bard will read out the response for you. It’ll even go a step further and evaluate whether there’s content across the web to back up that answer.
Another cool feature is shared conversations. If someone sends you a Bard chat through a public link, you can jump right in and keep the conversation going. You can ask more questions, explore new ideas, or simply use it as a starting point for further discussions. It’s a great way to collaborate and brainstorm with others.
And lastly, Google has expanded access to existing English language features. This means that more languages now have access to cool features like uploading images with Lens, getting Search images in responses, and even modifying Bard’s responses.
These awesome updates were made possible thanks to the PaLM 2 model. Google is really pushing the boundaries with Bard, and we’re excited to see how it continues to evolve and make our lives easier.
Intel has some exciting news for AI enthusiasts. Their new chip, set to launch in December, will allow you to run generative AI chatbots directly on your laptop, without relying on the computing power of cloud data centers. How cool is that?
This breakthrough has been made possible thanks to Intel’s upcoming “Meteor Lake” laptop chip, which includes innovative AI data-crunching features. And that’s not all. Intel is also releasing new software tools to support this capability, ensuring a seamless user experience.
But wait, there’s more. During a demonstration, Intel showed off laptops that could not only generate a song in the style of Taylor Swift but also answer questions in a conversational manner, all without needing an internet connection. Talk about flexibility and convenience!
And it doesn’t stop there. If you’re a fan of Microsoft’s Copilot AI assistant, you’ll be thrilled to know that it will be compatible with Intel-based PCs. This means you can enjoy the power and convenience of Copilot right on your own laptop.
With Intel’s latest innovations, the world of AI is becoming more accessible and versatile than ever before. Get ready to unleash the potential of AI right from the comfort of your own laptop. Exciting times lie ahead!
So, check this out. DeepMind, the artificial intelligence wunderkind from Google, has come up with a pretty amazing system called AlphaMissense. And get this—it’s designed to predict genetic diseases! How cool is that?
So, here’s how it works. This AI system looks at the letters in our DNA and figures out if they’ll produce the correct shape. If not, it flags them as potentially disease-causing. It’s like having a super-smart DNA inspector on the case!
Now, here’s the thing. Up until now, genetic disease hunters didn’t have a whole lot of knowledge about which parts of our DNA could lead to diseases. They had to sift through billions of chemical building blocks to find any clues. Talk about a needle in a haystack!
But with AlphaMissense, things are changing. DeepMind’s system has been able to classify a whopping 89% of the letter changes, or mutations, as either benign or disease-causing. That’s a huge leap from the mere 0.1% they could classify before. This is a game-changer!
Imagine the possibilities. This AI breakthrough could revolutionize the field of genetics and help researchers better understand and develop treatments for genetic diseases. Who knows what other mysteries we’ll uncover with the help of DeepMind’s brilliant algorithms? It’s an exciting time to be alive!
OpenAI recently introduced their latest text-to-image model, DALL·E 3, that has the ability to transform detailed requests into accurate and highly precise images. This new model is native to ChatGPT, allowing users to utilize ChatGPT to create tailored and specific prompts for DALL·E 3. And the best part is, if the generated image isn’t exactly what you envisioned, you can easily ask ChatGPT to make adjustments.
Comparing DALL·E 3 to its predecessor, DALL·E 2, the improvements are quite remarkable. With the same prompt, DALL·E 3 consistently produces significantly better results, as demonstrated in a visual comparison of an expressive oil painting of a basketball player dunking, depicted as an explosion of a nebula.
OpenAI has made it a priority to ensure that DALL·E 3 doesn’t generate images with violent, adult, or hateful content. They’ve implemented measures to prevent the model from creating such objectionable outputs. Additionally, DALL·E 3 has been designed to decline requests for images in the style of living artists. Creators also have the option to exclude their images from being used in the training of OpenAI’s future image generation models.
Currently, DALL·E 3 is in the research preview stage and will be made available to ChatGPT Plus and Enterprise customers in October through the API. Later this fall, it will also be introduced in Labs. So, exciting things are coming up for those who are eager to dive into the world of text-to-image generation with DALL·E 3!
Amazon recently revealed some exciting news at its annual devices event. They are introducing generative AI to both Alexa and Fire TV, offering enhanced experiences for users.
One notable update is the implementation of a new generative AI model within the Echo family of devices. This model has been specifically designed to optimize voice interactions. It delves deeper into conversational experiences by considering not just the spoken words, but also factors like body language, eye contact, and gestures. This holistic approach is set to make interactions with Alexa even more powerful and intuitive.
In addition to the Echo devices, Amazon is also bringing generative AI updates to Fire TV’s voice search functionality. This means that interacting with Alexa on Fire TV will become more conversational, allowing users to discover new content based on specific preferences and requests.
These AI updates pave the way for more natural and seamless interactions with Amazon’s smart devices. By incorporating body language and context, Alexa will be able to better understand user needs and preferences, leading to a more personalized and intuitive user experience.
With these generative AI advancements, Amazon is taking a significant step towards integrating AI technology into our everyday lives, making our interactions with Alexa and Fire TV more conversational, engaging, and enjoyable.
So, we’ve been hearing a lot about these self-supervised language models (LLMs), right? Well, turns out they’re not just impressive at predictive tasks, they also have some serious compression skills!
In this pretty interesting research, DeepMind and Meta, both part of the big G, decided to delve into the compression abilities of LLMs. They wanted to understand the relationship between compression and prediction. And guess what? They found out that these foundation models, which are primarily trained on text, are like the Jack-of-all-trades when it comes to compression.
You see, these LLMs have this amazing ability to learn within the context of the text they’re fed. And this in-context learning superpower makes them pretty effective at compressing all sorts of things. Take for example Chinchilla 70B (cute name, by the way), it achieves compression rates of 43.4% on ImageNet patches and 16.4% on LibriSpeech samples. And those numbers are even better than what you’d get with domain-specific compressors like PNG or FLAC.
So, what’s the big takeaway here? Well, next time you need something compressed, maybe you should consider calling up one of these LLMs. They might just surprise you with their incredible compression skills!
So, get this—Meta, you know, the company behind all those fancy virtual reality headsets and whatnot, has some pretty cool plans up their sleeves. They’re actually working on creating a whole bunch of chatbots that are specifically designed to connect with younger users. And get this—they’re not just your average run-of-the-mill bots either. Oh no, these chatbots are gonna have personalities. Yeah, you heard me right, personas!
The idea behind it is to make chatting with these bots a lot more exciting and engaging for the young crowd. And honestly, who doesn’t love a little extra color and sass, right? I mean, sure, you’ll be able to chat with them about all sorts of stuff, but these bots are gonna have some serious attitude.
But it doesn’t stop there! Meta’s got big plans for their chatbot army. They’re not only creating personalities for the young ones, but they’re even putting together some bots for celebrities. That means you might actually get to have a little tête-à-tête with your favorite star! How cool is that?
And last but not least, Meta’s also developing productivity-focused chatbots. You know, the kind that’ll help you with coding, managing tasks, and basically being all-around handy. So, whether you’re looking for some entertainment or a helping hand, it sounds like Meta’s got you covered with their upcoming sassy chatbots. Exciting times ahead, my friends!
Have you ever wished you could extend the context sizes of your pre-trained LLMs without spending a fortune on computation costs? Well, new research has come to the rescue with LongLoRA, an ultra-efficient fine-tuning method. This method allows you to increase the context length of your LLMs without the need for an abundance of time or powerful GPU resources.
Normally, training LLMs with longer context sizes can be a time-consuming and resource-heavy process. Just imagine, extending the context length from 2048 to 8192 could increase computational costs by a whopping 16 times! This is where LongLoRA steps in to save the day. It achieves efficiency in two key ways:
Firstly, it utilizes sparse local attention instead of dense global attention, which is optional during inference time. This clever technique helps to significantly reduce computational requirements.
Secondly, LongLoRA incorporates LoRA (Low-Rank Adaptation) for context extension. By leveraging LoRA, the method achieves remarkable performance while keeping the original model architectures intact.
The results speak for themselves. LongLoRA has demonstrated impressive performance across a range of tasks using LLaMA-2 models, from 7B/13B to 70B. Notably, it was able to extend LLaMA-2 7B from a 4k context to an impressive 100k, and LLaMA-2 70B to 32k – all on a single 8x A100 machine. This makes LongLoRA both easy to use and incredibly practical.
So, if you’re looking to maximize the capabilities of your LLMs without breaking the bank, LongLoRA might just be the solution you’ve been waiting for.
Exciting news! OpenAI is taking conversational AI to a whole new level with its latest update to ChatGPT. They’re introducing voice and image capabilities, making it possible for users to have voice conversations and show images to ChatGPT. This means that interacting with ChatGPT just got a whole lot more intuitive and versatile.
With this new feature, users can now engage in live conversations about landmarks. Imagine being able to ask ChatGPT about a famous monument and getting all the information you need just by describing it to the model. Additionally, if you’re stuck on what to cook for dinner, you can simply show pictures of the ingredients in your fridge to ChatGPT and receive recipe suggestions. How convenient is that?
But it doesn’t stop there! OpenAI is also allowing users to share photos to receive math problem hints. So, if you’re struggling with a tricky equation, just snap a quick picture and let ChatGPT assist you. This capability will surely come in handy for students and anyone who loves a bit of mathematical challenge.
The voice and image capabilities will be gradually rolled out to Plus and Enterprise users over the next couple of weeks. Voice will be available on both iOS and Android, while images will be supported on all platforms.
What’s more, ChatGPT can now understand and reason about images, including photos, screenshots, and even text-containing documents. And if you need to discuss multiple images, they’ve got you covered. OpenAI has also introduced a new drawing tool that can help guide you.
Get ready to experience ChatGPT like never before. The era of voice and image conversations has arrived!
Getty Images has recently introduced a cool new AI art tool called Generative AI. What it does is pretty remarkable – it uses an AI model provided by NVIDIA to transform text descriptions into actual images. Isn’t that mind-blowing?
But here’s the interesting part: Getty Images has made sure that this tool is not only technologically advanced but also “commercially safer” than other similar solutions out there. They have put in some measures to prevent disinformation and copyright infringement. That’s definitely a smart move considering how important it is to protect copyrights and avoid spreading false information.
Getty Images is also being fair to the talented artists out there. They will compensate the contributors whose work is used to train the AI generator and even share the revenues generated from the tool. Now that’s a great way to support the creative community!
If you’re eager to try out Generative AI, you can access it on Getty’s website. And hey, it’s not just limited to their website. You can also integrate the tool into apps and websites through an API. That’s pretty cool because it means more people can get to enjoy its creative possibilities.
Getty Images isn’t the only company exploring ethical approaches to generative AI. Bria and Shutterstock are also delving into this fascinating field. It’s exciting to see companies taking responsibility and making sure AI is used in a way that benefits everyone ethically and legally.
Colossal-AI recently launched an exciting new product called Colossal-LLaMA-2. This domain-specific language model solution is making waves in the industry, and for good reason. One of the standout features of Colossal-LLaMA-2 is that it is open-source and completely commercial-free. This means that users can take advantage of all its capabilities without having to worry about any hidden costs or subscriptions.
What makes Colossal-LLaMA-2 even more impressive is that it achieves outstanding results while using a relatively small amount of data and training time. This not only saves users valuable resources but also lowers costs significantly. It’s no wonder that this model has quickly gained attention and recognition.
In addition to its cost-effectiveness, the Chinese version of LLaMA-2 has proven to be a top performer in various evaluation benchmarks. Colossal-AI has made some noteworthy improvements with this release, including vocabulary expansion, a data cleaning system, and a multi-stage pre-training scheme. These enhancements aim to enhance the model’s abilities in both Chinese and English.
With Colossal-LLaMA-2, users can rely on a high-quality, commercial-free language model solution that delivers impressive results. Whether you’re working with Chinese or English, this model is sure to provide you with the accuracy and performance you need.
Hey there! Big news in the world of artificial intelligence! OpenAI, the AI research company, is apparently considering selling shares, and this could potentially skyrocket its valuation. The current estimation is that it could jump from $29 billion all the way up to a staggering $80 billion to $90 billion range! That’s a massive leap!
Moving on to another interesting tidbit, it seems like Jony Ive, the former design genius behind Apple, and Sam Altman, the CEO of OpenAI, have been having some intriguing conversations. Apparently, they’ve been discussing the possibility of creating a brand new AI hardware device. We don’t have all the details yet, but it’s exciting to ponder what kind of innovative hardware they could come up with for the AI era.
It’s always fascinating to see how AI continues to evolve and captivate the minds of brilliant individuals. We’ll definitely be keeping an eye on OpenAI and any updates regarding their potential share sale and the AI hardware device discussions between Jony Ive and Sam Altman. Who knows what revolutionary advancements lie ahead in the world of AI? It’s certainly an exciting time!
Have you heard the news? Vectara has just launched Boomerang, the latest and greatest LLM that is completely changing the game when it comes to GenAI accuracy. What’s so special about it, you ask? Well, it’s outperforming all the big-name competitors out there and is setting a whole new benchmark for Grounded Generative AI in the business world.
Boomerang is part of Vectara’s GenAI platform and is a next-gen neural information retrieval model that is taking things to the next level. But what exactly does it do better than the rest? For starters, it surpasses Cohere in terms of benchmark performance and matches OpenAI on certain metrics. But where it truly shines is in multilingual benchmarks. Boomerang has taken it upon itself to prioritize security, making sure that there is reduced bias, copyright concerns, and what they call “hallucinations” in AI-generated content.
But that’s not all! Boomerang goes above and beyond by offering cross-lingual support for hundreds of languages and dialects. It also improves prompt understanding, which ultimately leads to more accurate and faster responses. So, if you’re looking for the best of the best when it comes to GenAI accuracy, Boomerang is the way to go.
Google recently celebrated its 25th birthday and took a moment to reflect on its impressive legacy in the field of artificial intelligence. It all began back in 2001 when Google started using a machine learning algorithm to suggest improved spellings for web searches. Quite a humble beginning, but little did they know that it would pave the way for groundbreaking innovations.
One significant moment that stands out is the introduction of PaLM 2 and Gemini in 2023. These advanced models hold great promise and are expected to drive Google’s AI advancements for the next 25 years. With PaLM 2 and Gemini, Google aims to push the boundaries of what artificial intelligence can achieve.
Looking back on their 25-year journey, it’s clear that Google’s commitment to AI has been constant. From its early days of improving web search spellings to now revolutionizing various industries with cutting-edge technology, Google has demonstrated its passion for pushing the limits.
As we move into the future, there is no doubt that Google’s legacy in AI will continue to guide their innovative pursuits. With each passing year, we can anticipate even greater advancements that will shape the way we interact with technology and enhance our lives.
Google’s 25-year AI legacy sets the stage for an exciting future, as the company remains at the forefront of artificial intelligence research and development.
Have you heard about NVIDIA’s latest software called TensorRT-LLM? It’s designed to revolutionize LLM inference on H100 GPUs, delivering a whopping 8x performance boost. How impressive is that?
This software is packed with incredible features. It comes with optimized kernels, pre- and post-processing steps, and even multi-GPU/multi-node communication primitives to ensure top-notch performance. What’s really cool is that developers can experiment with new LLMs without needing deep knowledge of C++ or NVIDIA CUDA. So, it’s user-friendly and accessible to a wider range of developers.
To make things even better, TensorRT-LLM provides an open-source modular Python API, making customization and extensibility a breeze. Developers can easily tailor the software to their needs.
Now, if you thought that was it, there’s more. This software allows users to quantize models to FP8 format, which improves memory utilization. This is great news for those looking to optimize their memory usage and overall performance.
Currently, TensorRT-LLM is available in early access and will soon be integrated into the NVIDIA NeMo framework. If you’re interested, you can apply for access through the NVIDIA Developer Program, with a focus on enterprise-grade AI applications. So, keep an eye out for it!
NVIDIA is definitely pushing the boundaries of LLM deployment performance with TensorRT-LLM. It’s truly an exciting development in the world of AI.
Today, we have some exciting news from Google DeepMind. They have introduced a fascinating concept called Optimization by PROmpting, or OPRO for short. What is OPRO, you ask? Well, it’s all about using language models as optimizers.
Let me break it down for you. DeepMind’s language models are now being trained to generate new solutions based on a given problem and previously discovered solutions. So, when faced with an optimization problem, these models are able to describe it in natural language and come up with innovative solutions.
But how does it perform? The results are pretty impressive. OPRO has been applied to various tasks like linear regression, traveling salesman problems, and prompt optimization tasks. And guess what? The performance of the prompts optimized by OPRO surpassed human-designed prompts. In fact, it outperformed them by up to 8% on GSM8K and a whopping 50% on Big-Bench Hard tasks.
This breakthrough has the potential to revolutionize optimization methods. By leveraging the power of language models, DeepMind is exploring new avenues for solving complex problems more effectively. Exciting times lie ahead as language models continue to evolve and enhance our problem-solving abilities.
Meta has some ambitious plans in the works! Word on the street is that they’re gearing up to give OpenAI’s GPT-4 a run for its money. How, you ask? Well, by training their very own chatbot model that’ll give GPT-4 a serious run for its money. Meta means business.
Now, here’s the interesting part. Meta’s going all out for this venture. They’re snatching up Nvidia H100 AI-training chips, which means they won’t have to rely on Microsoft’s Azure cloud platform for training their chatbot. Smart move, if you ask me. They’re also working on expanding their data centers to beef up their chatbot’s capabilities. Talk about leveling up!
But wait, there’s more. CEO Mark Zuckerberg has big dreams for this chatbot model. He wants it to be available to companies for free so that they can create their own AI tools. Imagine the possibilities! With Meta’s model, companies can start developing AI tools that can mimic human expressions. That’s pretty darn cool.
So, keep your eye on Meta, folks. They’re going head-to-head with OpenAI, and who knows what kind of AI magic they’ll whip up next. It’s an exciting time for AI enthusiasts, that’s for sure!
Google is taking a big leap towards responsible AI development with the launch of its Digital Futures Project. They’re putting their money where their mouth is by establishing a $20 million Google.org fund, which will provide grants to top-notch think tanks and academic institutions worldwide.
The aim of this project is to empower researchers who are tackling the challenging task of developing AI responsibly. Google wants to foster a global conversation and organize events that bring together experts to discuss and debate policy solutions for responsible AI.
They’re starting off strong with some impressive names as the inaugural grantees of the Digital Futures Fund. Institutions like the Aspen Institute, Brookings Institution, Carnegie Endowment for International Peace, the Center for a New American Security, the Institute for Security and Technology, and SeedAI are among the fortunate recipients. And this is just the beginning. Google plans to support institutions from different corners of the globe.
With this initiative, Google is sending a clear message: they’re serious about making sure AI development is done the right way. By investing in research, organizing events, and supporting global experts, they’re paving the way for responsible AI that benefits society as a whole.
So, get this: Microsoft, MIT, and Google have teamed up for a groundbreaking project. They’ve taken the entire Project Gutenberg Collection and transformed it into audiobooks! Yeah, you heard that right. Now you can listen to thousands of free and open audiobooks, all thanks to some serious AI power.
But here’s the really cool part: they didn’t just stop at converting the texts into boring, monotonous readings. No, sir! They utilized the latest advancements in neural text-to-speech technology to give these audiobooks a real voice. And not just any voice, mind you. The quality of the voice acting is exceptional. It’s like having a professional narrator whispering the words in your ear.
But that’s not all. They’ve also created a system that lets you customize the audiobook experience. You can adjust the speaking speed and style to suit your preferences. Want a more emotional reading? No problem. They’ve got you covered. And get this – you can even find a voice that matches your desired sound by providing just a small sample of audio. How cool is that?
With this collaboration, the possibilities for audiobooks are truly endless. So, if you’re a bookworm who loves to listen on the go, or if you just want to immerse yourself in a great story, these AI-powered audiobooks are definitely worth checking out. Trust me, you won’t be disappointed.
Today, there is a huge demand for talent in Generative AI. It’s no surprise that the number of companies looking for “Generative AI” expertise in their job postings is soaring.
Tech giants like Amazon, Nvidia, Microsoft, and Google are leading the way in hiring GenAI talent. But they’re not alone! Big banks such as Citigroup and CapitalOne are also jumping on the GenAI bandwagon.
Interestingly, the technology sector is the top industry seeking GenAI experts, followed by finance and healthcare. Sectors like real estate, basic materials, and energy, on the other hand, haven’t been as enthusiastic about GenAI.
What’s even more intriguing is the salary companies are willing to pay for GenAI talent. Among all the technical skills and technologies being tracked, jobs mentioning “Generative AI” or “LLMs” had the highest average base salary at a whopping $200,837 per year.
Clearly, Generative AI is a hot field right now. The demand for talent is exploding, and companies are willing to pay top dollar to attract the best minds in the industry. So, if you’re interested in GenAI, there’s no better time to jump on board!
So listen up, folks! You might not have noticed, but Apple is sneaking in some AI wizardry into their snazzy new iPhones and watches. They’re slyly improving the basic functions of their gadgets using artificial intelligence. How clever!
At their developer conference, Apple didn’t shout about AI from the rooftops, but behind the scenes, their tech wizards have been weaving AI magic into their core software products. Sneaky, right?
Now, let me tell you about the cool features they’ve cooked up. First up, they’ve got these new-fangled semiconductor designs that power AI functions. With these babies, you can expect better call quality and sharper image capture on your new iPhone. Say goodbye to those pixelated photos!
But wait, there’s more! Apple’s watches are getting a boost too, thanks to their spanking new chip. It’s got a snazzy four-core “Neural Engine” that makes Siri even smarter. Rumor has it that Siri’s accuracy has shot up by a cool 25%. Impressive, right? Plus, this new chip opens up exciting new ways for you to interact with your trusty watch.
Oh, and here’s a neat trick: the new iPhone is so clever, it can automatically recognize the people in your photos. No more scrolling through hundreds of pics just to find that one shot of grandma blowing out her birthday candles. Apple’s got you covered!
So, while Apple might not be making some big fuss about AI, they’re certainly weaving their magic behind the scenes. Who knows what other tricks they’ve got up their sleeve?
Have you heard about Salesforce’s latest offering? They have introduced Einstein Copilot Studio, a tool that allows customers to customize their AI offerings. It’s a game-changer!
With the prompt builder, customers have the freedom to add their own custom prompts for their products or brands. Imagine being able to tailor the AI experience specifically to your business needs! This will undoubtedly give you an edge in the market and make your brand stand out.
But it doesn’t stop there. The skills builder feature is another exciting element of Copilot Studio. It enables companies to add actions to prompts, such as competitor analysis or objection handling. This means that you can make your AI not only smart but also incredibly useful, addressing various aspects of your business.
And if you’re wondering about the technical side of things, the model builder is here to save the day. It allows customers to bring their own models or even use supported third-party offerings. So, whether you have an existing model you want to integrate or prefer a ready-made solution, Salesforce has got you covered.
Now, I know what you might be thinking – what about biases and inappropriate responses? Well, Salesforce is one step ahead. They are actively working on a system called “the Einstein Trust Layer” to tackle those concerns. They understand the importance of fairness and ensuring that the AI delivers appropriate responses without any biases.
All in all, Salesforce’s Einstein Copilot Studio is a great tool for anyone looking to harness the power of AI in a customized and tailored way. It’s all about empowering you to create an AI experience that truly represents your brand and meets the specific needs of your business.
NExT-GPT is a groundbreaking AI system that is revolutionizing the world of human-like AI research. This multimodal language model is pretty impressive because it can comprehend and create content in different forms like text, images, videos, and audio. This means it’s not limited to just one way of understanding and generating information, but rather has a wide range of options at its disposal.
What’s even more fascinating about NExT-GPT is how it builds on existing models. By utilizing pre-trained encoders and decoders, it minimizes the need for extensive parameter tuning. This not only saves time but also enhances efficiency.
But that’s not all – NExT-GPT goes beyond mere optimization. It introduces something called modality-switching instruction tuning (MosIT). This nifty feature allows the system to seamlessly switch between different modalities, adapting to the specific requirements of the task at hand. This adaptability adds a whole new layer of intelligence to the model.
To further enhance performance, NExT-GPT also comes with a meticulously curated dataset specifically designed to tackle complex cross-modal understanding. This dataset empowers the system to comprehend intricate relationships between different modalities and generate richer, more contextually relevant content.
All in all, NExT-GPT is a remarkable innovation in AI research, redefining the boundaries of what AI can achieve in terms of human-like understanding and generation of content across multiple modalities.
Meta AI has introduced Belebele, an impressive dataset that is capable of understanding a staggering 122 languages. This exciting development allows text models to be evaluated across a wide range of languages, including those with high, medium, and low-resource availability. By expanding the language coverage of natural language understanding benchmarks, Belebele opens up new possibilities for language comprehension research.
The Belebele dataset itself is comprised of questions that are based on short passages extracted from the Flores-200 dataset. Each question offers four multiple-choice answers, designed to assess various levels of overall language comprehension. With this dataset, model performance can be directly compared across all languages. It has already been employed to evaluate both multilingual masked language models and large-scale language models.
Interestingly, the results have shown that smaller multilingual models tend to exhibit better performance in regards to comprehending multiple languages. This finding emphasizes the potential of compact yet versatile models when it comes to understanding diverse languages.
With Meta AI’s Belebele dataset, the field of language understanding and comprehension reaches new heights, enabling researchers to explore the intricacies of languages from around the world. This invaluable resource promises to drive further advancements in the development of language models and contribute to the overall progress of natural language processing technology.
Hey there! Guess what? Stability AI just came out with their 1st Japanese Vision-Language Model called Japanese InstructBLIP Alpha. It’s an absolute game-changer! This model is designed to generate textual descriptions for input images and even answer questions about them. How cool is that?
Let me fill you in on some of the details. This incredible model is actually built upon the Japanese StableLM Instruct Alpha 7B. It takes advantage of the InstructBLIP architecture. And you know what? It’s pretty darn good at what it does.
One of the things that makes this model stand out is its ability to accurately recognize Japan-specific objects. Plus, it can process text input like questions. Isn’t that amazing? It’s available on the Hugging Face Hub for inference and additional training. But here’s something important to note: it’s exclusively for research purposes.
Now, let’s talk about the applications. This nifty model can be used for so many things. It can power search engine functionality, provide scene descriptions, and even generate textual descriptions specifically for blind individuals. How awesome is that? Stability AI has really hit it out of the park with this one.
So, if you’re a researcher looking to explore the possibilities of vision-language models, this Japanese InstructBLIP Alpha is definitely something worth checking out. Go ahead, give it a spin!
So, here’s an interesting paper I came across that explores the relationship between transformers and support vector machines (SVMs). The authors establish a formal equivalence between the optimization geometry of self-attention in transformers and a hard-margin SVM problem.
What does this mean? Well, it turns out that when we optimize the attention layer of transformers, it actually converges towards an SVM solution. Specifically, this solution minimizes the nuclear norm of the combined parameter.
But that’s not all! The study goes on to prove that gradient descent also converges under certain conditions. And they even introduce a more general SVM equivalence for nonlinear prediction heads.
So what does all this mean in plain terms? It suggests that transformers can be thought of as a hierarchy of SVMs. These SVMs separate and select the most optimal tokens, allowing transformers to perform their magic.
This paper sheds some light on the inner workings of transformers, revealing that there’s a connection between transformers and SVMs. It’s fascinating to see how different machine learning techniques can be related and build upon each other.
Have you heard about Amazon’s latest breakthrough in AI-powered palm recognition? It’s called Amazon One, and it’s revolutionizing the way customers interact with everyday activities. This fast, convenient, and contactless device allows you to use the palm of your hand for various tasks, such as making payments, presenting loyalty cards, verifying your age, or even entering a venue. Yes, you heard that right – no need for your phone or wallet!
So, how does Amazon One work? It’s a combination of advanced technologies like generative AI, machine learning, cutting-edge biometrics, and optical engineering. These components come together to create a seamless and secure experience for users.
This innovative device is already making its way to over 500 Whole Foods Market stores and several other third-party locations, including travel retailers, sports and entertainment venues, convenience stores, and grocers. And it doesn’t stop there – Amazon One goes the extra mile by being able to detect fake hands and reject them, ensuring the highest level of security.
Here’s the impressive part – Amazon One has already been used over 3 million times with astounding accuracy of 99.9999%. This speaks volumes about its reliability and efficiency.
Imagine the convenience of leaving your phone and wallet at home while effortlessly completing various tasks with just the palm of your hand. Thanks to Amazon One, this futuristic concept is now a reality.
Did you know that Intel is really stepping up its game in the AI field? They’re not just focused on data center-based AI accelerators anymore. Nope, they’re expanding their reach and going after the AI opportunity in multiple ways.
According to Intel’s CEO, Pat Gelsinger, AI is going to become more accessible to everyday users. And why is that? Well, there are a few reasons. First, there are economic considerations. It’s getting cheaper and more affordable to incorporate AI technology into various products. Second, there are physical considerations. AI is moving closer to end-users because it’s no longer confined to giant data centers. And finally, there are privacy considerations. By having AI on devices, users have more control over their data.
So, how is Intel making this happen? They’re incorporating AI into their server CPUs, like the Sapphire Rapids. These CPUs have built-in AI accelerators specifically designed for inference tasks. But Intel doesn’t stop there. They’re also launching the Meteor Lake PC CPUs that have dedicated AI hardware. This means AI workloads can be accelerated directly on user devices. And with Intel’s dominant position in the CPU market, it’s a win-win situation. Software providers are more likely to support Intel’s AI hardware, making it even more appealing for everyday users like you and me.
It’s an exciting time for AI, and Intel is definitely making some bold moves to stay ahead of the game.
Hey there! If you’re itching to delve deeper into the world of artificial intelligence, I’ve got just the thing for you. There’s this super informative book out there called “AI Unraveled: Demystifying Frequently Asked Questions on Artificial Intelligence.” Trust me, it’s a gem.
What’s great about this book is that it’s all about making AI less cryptic and more accessible. If you’ve got questions about artificial intelligence swirling in your mind, this book is here to give you the answers you’re seeking. It’s like having a knowledgeable friend who’s always there to clarify things for you.
You might be wondering where you can get your hands on this literary marvel. Well, you’ve got options! This book is available at Apple, Google, and Amazon, so you can choose the platform that suits you best. Whether you love your Apple device, prefer to browse through Google Play, or are a dedicated Amazon enthusiast, you’re covered.
So, what are you waiting for? Go ahead and snatch up a copy of “AI Unraveled” today. With this book in your hands, you’ll be well on your way to unraveling the mysteries of artificial intelligence. Happy reading!
We covered a wide range of topics, including the evolution of search and language models, Amazon’s collaboration with Anthropic, Google’s personalized route suggestions in Google Maps, MIT’s Style2Fab AI tool, DeepMind’s AlphaMissense system for predicting genetic diseases, OpenAI’s DALL·E 3 text-to-image model, and Amazon’s generative AI updates for Alexa and Fire TV. We also discussed Meta’s chatbot personalities, the launch of Getty Images’ Generative AI art tool, Colossal-AI’s open-source language model, and the potential sale of shares by OpenAI. Additionally, we touched on NVIDIA’s TensorRT-LLM software, Google DeepMind’s language models as optimizers, and Apple’s integration of AI into their devices. Finally, we explored Salesforce’s Einstein platform, Meta AI’s Belebele dataset, Stability AI’s Japanese InstructBLIP Alpha model, and the expansion of AI by Intel. Don’t forget to check out “AI Unraveled,” a book that answers frequently asked questions about AI. Join us next time on AI Unraveled as we continue to demystify frequently asked questions on artificial intelligence and bring you the latest trends in AI, including ChatGPT advancements and the exciting collaboration between Google Brain and DeepMind. Stay informed, stay curious, and don’t forget to subscribe for more!
Keywords: LLMs, Search Engines, AI Rundown, AI Monthly, AI in September 2023, Future of Search, Language Models, AI developments, AI News, LLM vs. Search
Nvidia offices just got raided by french authorities
French antitrust authorities conducted a surprise raid on NVIDIA, suspecting anticompetitive practices in the graphics cards sector.
The raid was part of a larger investigation into the cloud computing market, focusing on major players like Amazon Web Services, Google Cloud, and Microsoft Azure.
The French Competition Authority emphasized that a raid doesn’t indicate guilt, but it is a significant move by the regulator.
Mark Zuckerberg unveils Meta’s new lifelike VR avatars
Podcaster Lex Fridman conducted what he calls the “first interview in the Metaverse” with Mark Zuckerberg, using their exceptionally realistic VR avatars created through extensive facial scans.
The 3D avatars, called Codec Avatars, map the user’s facial expressions in real-time and are more bandwidth efficient than transmitting video, according to Zuckerberg.
Although the currently displayed avatars use a state-of-the-art scanning process, Zuckerberg’s future plan is to generate similar quality avatars using a quick scan from users’ smartphones.
Tesla faces federal lawsuit over racial discrimination allegations
The US Equal Employment Opportunity Commission (EEOC) has sued Tesla, accusing it of violating the Civil Rights Act by creating a hostile work environment for its Black employees.
The lawsuit alleges rampant racial harassment including slurs, racial stereotypes, and offensive graffiti at Tesla’s Fremont, California, manufacturing facilities starting from May 2015.
The EEOC claims Tesla retaliated against Black employees who objected to discriminatory practices, and is seeking damages, reinstatement, and policy changes within Tesla.
Epic Games announces major layoffs and Bandcamp sale
Fortnite developer Epic Games is laying off 16% of its workforce, affecting approximately 830 employees.
The layoffs have been attributed to financial instability and major structural change to their economics, despite Fortnite’s growth and efforts to cut costs.
Severance packages, including six months of base pay plus health care coverage and other benefits, are being offered to those affected by the layoffs.
Meta AI: The new ChatGPT rival was trained on your posts
Meta’s new AI assistant, a potential rival to ChatGPT, is being trained using public posts from Facebook and Instagram.
Meta AI: ChatGPT’s Rival
Introduction to Meta AI: Launched at Meta Connect 2023, Meta AI aims to become a prominent assistant across platforms such as Instagram, WhatsApp, and Facebook.
Capabilities: Beyond just providing information like ChatGPT, it will perform tasks across various platforms and is set to integrate with products like the Ray-Ban Meta smart glasses and Quest 3.
Training on Your Posts
Data Source: The unique edge of Meta AI comes from its training on public posts from Facebook and Instagram, essentially learning from users’ informal content or “sh*tposts.”
Respecting Privacy: Meta takes care to not use private posts or messages for training, emphasizing the respect of user privacy.
Apple petitions Supreme Court to overturn a ruling which forces it to allow third-party payments in its App Store, contesting it’s unconstitutional.
Tim Cook, Apple CEO, asserts that the upcoming Vision Pro headset, fueled by app developers’ creativity, will revolutionize computing, work, communication, and our overall experience – comparing its potential impact to the iPhone’s.
SpaceX’s defense-oriented Starshield bags its first contract from the U.S. Space Force, enhancing satellite internet capacity and military resilience.
Footage from a Serve Robotics food delivery robot helped LA police to arrest and convict two attempted thieves, amid public surveillance concerns.
Steroid-promoting videos are increasing on TikTok, potentially endangering young men’s health and propagating harmful body image ideals, according to the CCDH.
Daniel Ek identifies three personas of Mark Zuckerberg: the young, egotistical founder, the controversial figure through Cambridge Analytica scandal, and the current, more responsible version.
Meta Platforms (META.O) Chief Executive Mark Zuckerberg on Wednesday rolled out new AI products for consumers, including bots that create photo-realistic images and smart glasses that answer questions, as well as an updated virtual-reality headset.
The European Union is examining alleged anticompetitive practices in chips used for artificial intelligence, a market that Nvidia (NVDA.O) dominates, Bloomberg News reported on Friday, citing people familiar with the matter.
Sex robots powered by futuristic AI algorithm will one day give humans the best sex of their lives, it has been sensationally claimed.
National Security Agency Director Army Gen. Paul M. Nakasone today announced the creation of a new entity to oversee the development and integration of artificial intelligence capabilities within U.S. national security systems.[4]
Emerging AI Innovations: September 29th, 2023
Can AI Create Another AI?
If yes, What you guys think, How Far Are we on achieving this? especially A general AI system can design and create another AI system by itself, without any human intervention or guidance.
I am exploring the possibility of self-replicating and self-improving AI systems, which could have profound implications for the future of humanity and civilization.”
I think that the answer is Yes, to a certain extent: AI can be involved in the creation of other AI systems. This process is typically known as automated machine learning (AutoML). AutoML involves using algorithms and computational power to automate the process of training and optimizing machine learning models.
In AutoML, AI systems, often referred to as “meta-learning” algorithms or “AI assistants,” are used to perform tasks like feature engineering, model selection, hyperparameter tuning, and even creating custom architectures. They aim to automate the labor-intensive aspects of machine learning, making it more accessible to a wider audience, including those without extensive expertise in data science.
However, it’s important to note that while AI can assist in the development and optimization of models, it does not possess the creative or conceptual abilities to create entirely novel AI algorithms or architectures from scratch. That level of innovation still relies on human researchers and engineers.
The NSA is establishing an “Artificial Intelligence Security Center”
The NSA is creating a new center focused on promoting secure AI development and defending U.S. advances from foreign adversaries aiming to co-opt the technology. (Source)
The AI Security Center
Aims to help spur the secure integration of AI capabilities.
Will develop best practices and risk management frameworks.
goal is to understand and combat threats to U.S. AI advances.
Motivations
The U.S. currently leads in AI, but the advantage is precarious.
Adversaries have long stolen intellectual property.
Agencies are adopting AI rapidly across missions.
I will work with industry, labs, and academia on priorities.
It comes after an NSA study showed the need to prioritize security.
Must understand AI vulnerabilities and counter-threats.
TL;DR: The NSA is establishing an AI Security Center to promote secure development and adoption of AI while defending U.S. progress from adversaries aiming to exploit the technology.
Google is expanding its AI-powered search experience to teenagers
Google’s AI-driven search experience, Search Generative Experience (SGE), is now accessible to teenagers between 13-17 in America. Entailments include a conversational mode for searches, which Google believes can help youngsters pose atypical questions to dig deeper.
Teen-friendly AI search
SGE introduces a conversational mode to Google Search, allowing users to ask questions and follow-ups in a more natural language.
To prevent harmful content from surfacing, Google has placed guardrails, providing stronger protections related to illegal and age-gated substances, or bullying.
Features and improving AI accuracy
Google is rolling out “About this result” to provide users with more context about the displayed content.
Google acknowledges and addresses any validation of false or offensive claims by the AI-powered response, ensuring to provide higher quality and more accurate responses.
It’s also using large language models to self-critique and rewrite draft responses on sensitive topics based on quality and safety principles.
SGE’s popularity and future plans
Since SGE’s introduction, it has found popularity, especially among younger users who prefer a conversational approach.
Google plans to expand SGE outside the U.S. to India and Japan and improve its services with support for videos, images, local info, and more.
It’s also experimenting with ads positioned next to the AI-generated responses.
Daily AI Update News from AWS, Meta, Google, Spotify, Zapier, Microsoft, Cohere, and Mayo Clinic
AWS has announced 5 major generative AI updates and innovations
Amazon Bedrock is now generally available.
Amazon Titan Embeddings is now generally available.
Meta’s Llama 2 is coming to Amazon Bedrock in the next few weeks.
New Amazon CodeWhisperer capability is coming soon, will allow customers to securely customize CodeWhisperer suggestions using their private code base to unlock new levels of developer productivity.
New generative BI authoring capabilities in Amazon QuickSight to help business analysts easily create and customize visuals using natural-language commands.
Meta introduces LLAMA 2 Long – In new research, Meta presents a series of long-context LLMs that support effective context windows of up to 32,768 tokens. The model series are built through continual pretraining from Llama 2 with longer training sequences and on a dataset where long texts are upsampled. Notably, the 70B variant can already surpass gpt-3.5-turbo-16k’s overall performance on a suite of long-context tasks. Check out the research for more interesting findings!
Google announces Google-Extended and expands access to SGE – Google-Extended will let web publishers control access to content on their site for training Google’s AI models. And Google has opened up access to SGE in Search Labs to more people, specifically teens in the U.S., so they too can benefit from generative AI’s helpful capabilities.
Spotify is adding auto-generated transcripts to millions of podcasts – The transcript feature will expand to more podcasters on Spotify and include time-synced text. In the future, creators could add media to transcripts– a useful feature if a creator is describing an image on the show, for example.
Zapier launches Canvas, an AI-powered flowchart tool – The tool aims to help its users plan and diagram their business-critical processes with a bit of AI to help them turn those processes into Zapier-based automations. Canvas is now in early access.
Microsoft opens AI Co-Innovation Lab in San Francisco to empower Bay Area startups – The lab’s main goal is to facilitate the transition from ideation to prototyping, providing companies with the resources and guidance they need to refine their AI-based concepts.
Cohere jumps into the fray of the AI chatbot race by releasing a new API – The Chat API with Retrieval-Augmented Generation (RAG) will allow third-party developers of other enterprises to build powerful chat applications based off Cohere’s proprietary generative LLM, Command.
Mayo Clinic to deploy and test Microsoft generative AI tools – Mayo Clinic is among the first healthcare organizations to deploy Microsoft 365 Copilot. It is testing the Early Access Program with hundreds of its clinical staff, doctors, and healthcare workers.
This project originally started out with a focus on easily generating embeddings from Llama2 and other llama_cpp (gguf) models and storing them in a database, all exposed via a convenient REST api. But since then, I’ve added a lot more functionality:
New endpoint for generating text completions (including specifying custom grammars, like JSON).
Get all the embeddings for an entire document– can be any kind of document (plaintext, PDFs, .doc/.docx, etc.) and it will do OCR on PDFs and images.
Submit an audio file (wav/mp3) and it uses whisper to transcribe it into text, then gets the embeddings for the text (after combining the transcription segments into complete sentences).
Integrates with my new vector similarity library (`pip install fast_vector_similarity`) to provide an “advanced” semantic search endpoint. This uses a 2-step process: first it uses FAISS to quickly narrow down the set of stored embeddings using cosine similarity, then it uses my vector similarity library to compute a bunch of more sophisticated (and computationally intensive) measures for the final ranking.
As a result, I changed the project name to Swiss Army Llama to reflect the new project goal: to be a one stop shop for all your local LLM needs so you can easily integrate this technology in your programming projects. As I think of more useful endpoints to add (I constantly get new feature ideas from my own separate projects– whenever I want to do something that isn’t covered yet, I add a new endpoint or option), I will continue growing the scope of the project. So let me know if there is some functionality that you think would be generally useful, or at least extremely useful for you!
A big part of what makes this project useful to me is the FastAPI backbone. Nothing beats a simple REST API with a well-documented Swagger page for ease and familiarity, especially for developers who aren’t familiar with LLMs. You can set this up in 1 minute on a fresh box using the docker TLDR commands, come back in 15 minutes, and it’s all set up with downloaded models and ready to do inference or get embeddings. It also lets you distribute the various pieces of your application on different machines connected over the internet.
Emerging AI Innovations: September 28th, 2023
Will LLMs replace Search?
They will evolve and blend together. – search offers freshness, relevancy, comprehensiveness – all gained in more than 29 years of industrial research[1] – large models offer phenomenal language understanding and generation, but sometimes lack verifiability, attributions, and relevancy – not surprisingly given they have been around for 6 years[2] In my view, we will see a rise of Language Search Models (shortly LSMs) where the best of the two words will emerge [1] The Anatomy of a Large-Scale Hypertextual Web Search Engine, 1998, Google [2] Attention Is All You Need, 2017, Google
Meta’s new exciting AI experiences & tools
Meta’s new AI features include an AI Assistant powered by Bing, It will provide real-time information and generate photorealistic images from text prompts. Meta used specialized datasets to train the AI to respond in a conversational and friendly tone. The first extension of the AI Assistant will be web search. The AI Assistant will be available in beta on WhatsApp, Messenger, and Instagram.
Introduced 28 AI personality chatbots based on celebrities, such as Tom Brady, Naomi Osaka, Mr. Beast, and more. These chatbots, accessible on platforms like WhatsApp, Messenger, and Instagram, provide topic-specific conversations but are currently text-based, with plans to introduce audio capabilities. These AI personalities were created using Llama 2. Meta aims to integrate Bing search functionality in the future. The chatbots’ animations are generated through AI techniques, offering a cohesive visual experience.
Launching AI Studio, a platform allowing businesses to build AI chatbots for Facebook, Instagram, and Messenger, initially focusing on Messenger for e-commerce and customer support apps. This toolkit will be available in alpha.
Gen AI stickers powered by Emu allow users to create unique stickers across its messaging apps. Users can type in their desired image descriptions, and Emu generates multiple sticker options in just a few seconds. Initially available to English-language users, this feature will roll out over the next month.
Introducing 2 new AI Instagram features, restyle and backdrop.Restyle allows users to transform the visual styles of their images by entering prompts like “watercolor” or more. While backdrop changes the background of photos using prompts.
Launches New-gen Ray-Ban smart glasses, in partnership with EssilorLuxottica, will feature improved audio and cameras, over 150 different custom frame and lens combinations. They’re lighter and more comfortable. Will enable livestream to Facebook or Instagram and use “Hey Meta” to engage with Meta AI assistant by voice.
Why does this matter?
Meta’s this move leads the race with technology heat-up. Its new AI-driven experiences and tools will enhance user interactions to the next level and open up numerous possibilities in communication and content creation.
Jony Ive and OpenAI’s plan to build the ‘iPhone of artificial intelligence’
Apple’s former chief design officer, Jony Ive, is reportedly in discussions with OpenAI to build the “iPhone of artificial intelligence,” aided by over $1 billion in funding from Softbank CEO Masayoshi Son.
If you want to stay ahead of the curve in AI and tech, look here first.
Design and Interaction
Design partnership with LoveFrom: OpenAI’s CEO, Sam Altman, plans to leverage Jony Ive’s design firm, LoveFrom, to develop OpenAI’s debut consumer device.
Aiming for natural interaction: Ive and Altman are focused on creating a device that offers a user experience mirroring the intuitiveness of the original iPhone’s touchscreen for AI interactions.
Influence and Funding
Inspired by iPhone’s success: The new device draws inspiration from the original iPhone, highlighting the transformative impact of touchscreen technology.
Softbank’s financial backing: Masayoshi Son is not only providing over $1 billion in funding but has also suggested involving chip design company Arm in the project.
Focus on Healthy Tech Consumption
Ive’s concerns on smartphone addiction: Jony Ive has historically voiced concerns over the addictive nature of smartphones.
Vision for screenless computing: The collaboration with OpenAI presents an opportunity for Ive to work on an interactive device less dependent on screens.
ChatGPT is back with internet browsing, It can now browse the internet to provide current & reliable information, along with direct links to sources. This update addresses feedback received since the browsing feature was launched in May. The model now follows robots.txt and identifies user agents to respect website preferences.
Currently available to Plus and Enterprise users, browsing will be expanded to all users soon.
To try it out, enable Browse in your beta features setting:
Click on ‘Profile & Settings’ > Select ‘Beta features’ > Toggle on ‘Browse with Bing’ > Choose Browse with Bing in the selector under GPT-4.
Why does this matter?
Users can now access up-to-date and trustworthy information with direct source links, making it a valuable tool for research, problem-solving, and knowledge acquisition. Also, they’re trying to make it a valuable resource for a broader audience.
The difference between AI creativity and human creativity, and how it is rapidly narrowing.
While many consider human creativity to be truly original and superior in results, it appears boundaries between AI-generated content and human creativity are becoming increasingly blurred. And it’s looking increasingly likely that AI may soon be at par with humans in creative content generation. Let’s look at a quick comparison between humans and ChatGPT to understand this:
Definition
Creativity in the context of content creation is the ability to produce original and innovative content that is engaging, informative, and valuable to the target audience. By this definition, both human and AI can create new content by logically combining existing knowledge based on the unique context and purpose. Both sources have also been verified to create valuable, informative, and helpful content.
However, humans may be more capable of abstract thinking that breaks free from typical creative patterns that AI is trained on. This is because they are influenced by diverse factors (like bias, flawed memory, multiple senses, emotions and more), apart from past knowledge and logic. As AI advances to better multimodal and experential learning, this difference may reduce greatly.
Basis of Creativity:
Humans- Our creativity often stems from a combination of stored knowledge, experiences, and innate cognitive processes. We remix, reframe, and reinterpret what we know to create something new.
GPT & AI- Similarly, AI models like GPT generate content based on patterns in their training data. They can combine and restructure this information in novel ways.
Initiation:
Notably, AI also lacks spontaneous initiation and requires prompting by humans as it does not have personal objectives or desires. The output quality largely depends on the input and can be manipulated easily, unlike other humans. This passive and dependent basis makes AI appear less creative, but that may just be because it is designed to depend on and adjust to human inputs.
Influence of Emotions and Identity:
Humans- Emotions and personal identity play a significant role in many creative endeavors. They influence tone, nuance, and depth in storytelling, poetry, and even in some journalistic pieces.
AI- AI can copy these influences easily based on patterns in the input data! It may not genuinely “feel” or “experience” emotions or identity, but that hardly makes a difference in the final output as long as it can recreate the influence of emotions.
Multimodal Learning and Experience:
Humans- Humans learn from their unique experiences due to their distinct life situations, and they use multiple senses to gain unique and holistic knowledge i.e. multimodal learning. ChatGPT can not do this as of now.
AI- Advancements in AI and IoT are leading to models that can process and generate content across multiple modalities (text, image, audio, video). This brings AI closer to the way humans perceive and interact with the world, allowing it to form unique multimodal experiences that it can apply to content creation.
Dynamic Learning and Memory:
Humans- Human memories are temporary and flawed as they fade or even vary over time. We also continuously learn, adapt, and change our perspectives based on new experiences and knowledge. We also perform research for new information. So, the creativity of an individual will naturally vary over time.
AI- Current models like GPT have static and permanent knowledge once trained. However, future models might incorporate dynamic learning, allowing them to adapt over time without explicit retraining. (covered in the previous point). This is already partly applicable as LLMs are being integrated with search engines and use researched online data.
Ethical Considerations:
Humans- Human creativity is often bound by ethical considerations, cultural sensitivities, and a sense of responsibility. Humans also understand the context in detail with higher accuracy due to multimodal experiences and personal understanding of the audience’s perspective (empathy).
AI- AI can be programmed to follow certain ethical guidelines, but the nuances and complexities of human ethics and an understanding of the audience’s perspective can be challenging to encapsulate and replicate fully. This may cause harmful outputs or restrict the creative abilities of the LLM.
Output quality:
Humans- It all comes down to the quality of creative outputs, and a skilled human writer with good language, research/knowledge, and reasoning skills can make highly creative outputs while maintaining accuracy and moderation. There is still the chance for human error or personal bias, which can possibly be corrected by the same writer through review and introspection.
AI- AI can also craft very creative and accurate outputs through existing knowledge or online research and typically tends to avoid bias or controversial outputs. However, there are cases where it may hallucinate and create false/harmful outputs or refuse to provide harmless outputs unless proper context and knowledge are manually fed. Since it cannot understand the reader’s perspective or user’s intent directly and does not have an understanding that the data it is consulting may be false, these errors will likely continue.
In the future, as AI becomes more advanced and incorporates more dynamic and multimodal learning as well as a better understanding of the audience, the gap between AI-generated content and human creativity might narrow further. What do you think?
Whose societal worth will AI hurt the most?
Now that we know AI’s capabilities by rigorously testing chatgpt4 … I’ve been pondering how it will impact the future. Let’s be real, it won’t replace everyone. If you got by in life by being good looking, AI won’t replace you. I don’t care how hot that silicone AI bot will be; a hot human will always be preferable. Who will it replace? I think the nerd … that person who struggled in highschool but who found their way in the world by getting a job that required brainpower and paid well. Not amazingly well, but like a few hundred k.
Nerds were the bottom rung of society for thousands of years. When society consisted of vikings and warriors we were nobodies. When society consisted of peasants and knights we were nobodies. When society consisted of tradesmen and seafarers we were nobodies. Remember that for the vast vast majority of history, simply being able to read and write was rare. There was little use for that skill so no one bothered to learn it.
Note I’m not talking about the rare geniuses like the Teslas, Newtons, Einsteins … I mean your run of the mill 120-140IQ nerd who could do an intellectually hard job well. The people who now become lawyers, radiologists, doctors, MBAs, accountants, reporters, programmers, engineers … We only started to gain worth when our brainpower and patience for tedium had value, during a brief period in the last century or, at most, two centuries. Now Sam Altman, Greg Brockman and Ilya Sutskever (all three of whom are clearly one of us) destroyed that one path we had to worth with their invention. Oh the irony.
Of course these are just my thoughts and I wanted to get other opinions but that’s where I’m leaning. Will AI return us nerds to a pre-modern age era where we were among the lowest group in society, because there was little use for our brainpower?
Comments:
1- Lawyers, I worked in complex litigation for a long time. Very few cases create new law. It’s 95% copy and pasting and paralegals do most of the legal research to find primary and secondary sources to support the arguments made. AI can do that in no time.
2- In the long run I think everyone will have to reconsider individualism and accept that we just aren’t that special and can be replicated in almost every aspect (given enough time with AI progression)
3- In theory, everyone that does work on a PC that doesn’t involve physical labor. AGI is going to be so powerful that it will basically be like talking to the smartest person you ever met that can answer almost anything it’s trained on almost instantly. Our only hope is that the compute cost to operate it is so high that it’s cheaper to employ smart people instead.
Getting Emotional with LLMs Can increase Performance by 115% (Case Study)
This research was a real eye-opener. Conducted by Microsoft, the study investigated the impact of appending emotional cues to the end of prompts, such as “this is crucial for my career” or “make sure you’re certain.” They coined this technique as EmotionPrompt. What’s astonishing is the significant boost in accuracy they observed—up to 115% in some cases! Human evaluators also gave higher ratings to responses generated with EmotionPrompt. What I absolutely love about this is its ease of implementation—you can effortlessly integrate custom instructions into ChatGPT. We’ve compiled a summary of this groundbreaking paper. Feel free to check it out here. For those interested in diving deeper, here’s the link to the full paper.
What is EmotionPrompt?
The groundbreaking study, known as “EmotionPrompt,” delves into the impact of emotional stimuli on LLMs. Instead of merely asking the model factual questions, the study introduced emotional context. For example, instead of asking, “Is this statement true or false?”, the prompt would be, “Is this true or false? This is crucial for my career.”
Why This Matters
Enhanced Performance
Firstly, the study found that adding emotional context improves the model’s performance. Imagine you’re a business owner who needs to analyze large sets of customer feedback. An LLM with emotional intelligence can do this more accurately, akin to a heightened level of focus.
Increased Truthfulness and Informativeness
The study also revealed that emotionally intelligent LLMs are more truthful and informative. This is particularly beneficial in sectors that require factual accuracy, such as healthcare or law.
Greater Stability
Interestingly, these models also showed less sensitivity to changes in their settings, making them more reliable. In technical terms, they are less sensitive to “temperature” adjustments, which means you can count on consistent performance.
The research team initially compiled a roster of emotional triggers for experimental use, drawing upon three foundational theories in psychology: Self-Monitoring, Social Cognitive Theory, and Cognitive Emotion Regulation Theory.
Key Takeaways:
EmotionPrompt consistently received higher ratings across all evaluation metrics.
Specifically, in terms of performance, EmotionPrompt realized a relative gain of 1.0 or more (equating to a 20% increase) in nearly a third of the tasks.
Only on two occasions did EmotionPrompt fall short.
In a comparative analysis of poem composition, EmotionPrompt’s poem was deemed more creative.
EmotionPrompt led to a 19% uptick in truthfulness.
The human study corroborates the quantitative data, underscoring EmotionPrompt’s practical relevance and user resonance.
Concluding Remarks
Final Insights from the Study:
Merging multiple emotional triggers yielded marginal or no additional benefits.
The potency of emotional stimuli is task-dependent.
Larger LLMs stand to gain more from EmotionPrompt.
As the temperature setting escalates, so does the relative gain.
OpenAI partners with WHOOP to launch WHOOP Coach, an advanced-gen AI feature for wearables. It uses OpenAI’s GPT-4 system to provide personalized recommendations & guidance for health and fitness. The feature analyzes WHOOP data, sports science, and individual body information to generate personalized answers.
Cloudflare launched new AI tools to help customers build, deploy, and run AI models at the network edge. The first tool, Workers AI, allows customers to access nearby GPUs on a pay-as-you-go basis. Another tool, Vectorize, provides a vector database to store mathematical representations of data. The third tool, AI Gateway, offers metrics to help customers manage the costs of running AI apps.
Microsoft & Mercy partners for Clinician Empowerment with Gen AI. The partnership allowed Mercy to make real-time clinical decisions & improve patient care. They are exploring over four dozen uses of AI and plan to launch multiple new AI use cases by next year to enhance patient and co-worker experiences.
Adobe has officially launched Photoshop on the web, a simplified online version of its popular desktop photo editing app. The web version includes AI tools such as Generative Fill and Generative Expand, powered by Adobe’s Firefly generative AI model. These tools allow users to manipulate images using text-based descriptions in over 100 languages.
Microsoft plans to use nuclear energy to power its AI data centers – The company is recruiting a “principal program manager for nuclear technology” to evaluate the feasibility of using nuclear energy to support the energy demands of hosting AI models. The company sees nuclear energy as a viable option to address the escalating energy demand of running AI models like ChatGPT.
Emerging AI Innovations: September 27th, 2023
CIA’s ChatGPT Revolution, OpenAI’s Valuation Surge & AI Milestones from Google to VideoDirectorGPT
Unravel the CIA’s new ChatGPT-like technology and OpenAI’s valuation leap to $90B. Dive deep into Boomerang, the LLM reshaping GenAI accuracy, and reflect on Google’s quarter-century AI journey. Explore the future with open-source AI, tools for perfecting prompt quality, and the magic of VideoDirectorGPT transforming text into multi-scene visuals. Discover how Colossal-AI is redefining ad-free experiences.
The Central Intelligence Agency (CIA) has some exciting news to share. They are in the process of developing a ChatGPT-style AI that will be used within the US intelligence community. This AI aims to revolutionize data analysis and intelligence gathering efforts. It’s a big step forward for the intelligence community, which includes 18 different agencies such as the CIA, NSA, FBI, and various military offices. So, what exactly will this AI do? Well, it’s a large language model (LLM) chatbot that has been specifically designed to provide summaries of open-source materials and citations. But that’s not all – it can also engage in conversations with users, potentially answering questions and providing additional information. The goal is to help analysts efficiently parse through massive amounts of data and have interactive conversations with the AI, expanding their capacity for collecting and processing information. One of the most exciting aspects of this AI is its potential for unlimited data collection. Randy Nixon, the director of the CIA’s Open-Source Enterprise, emphasized this point, noting that the technology allows for continuous growth, with the only limitation being budgetary considerations. This means that the intelligence community can gather as much data as they need, without any constraints. Nixon also highlighted the continuous evolution of technology in the intelligence field. From traditional media sources like newspapers and radio, we have now moved into the era of data-driven approaches. He stressed the importance of adapting and effectively harnessing the capabilities of AI to stay ahead in the game. Interestingly, this announcement coincides with China’s own aspirations to lead in AI technology globally by the end of the decade. China has already introduced strict regulations for AI services’ security assessments, which could potentially impact technological advancements in this field.
The Central Intelligence Agency (CIA) has confirmed its development of a ChatGPT-style AI for use within the US intelligence community, aiming to enhance data analysis and intelligence gathering efforts.
Source Randy Nixon, director of the CIA’s Open-Source Enterprise, described this AI project as a significant technological advancement for the intelligence community, which comprises 18 agencies, including the CIA, NSA, FBI, and various military offices. The large language model (LLM) chatbot is designed to provide summaries of open-source materials and citations, as well as engage in conversations with users, potentially answering questions and providing information. The goal is to enable analysts to parse vast amounts of data efficiently and have interactive conversations with the AI, expanding the capacity for collecting and processing information.
Unlimited Data Collection: Randy Nixon emphasized the potential for limitless data collection with such a system, noting that the technology allows for continuous growth, constrained only by budgetary considerations.
Technological Evolution: Nixon highlighted the continuous evolution of technology in the intelligence field, from traditional media like newspapers and radio to modern data-driven approaches, stressing the need to adapt and harness AI’s capabilities effectively.
China’s AI Ambitions: The announcement coincides with China’s aspirations to lead in AI technology globally by the end of the decade. China has introduced stringent regulations for AI services’ security assessments, potentially impacting technological advancements.
CIA’s Tech Recruitment Efforts: The CIA has been actively seeking tech talent and partnerships with the private sector to leverage AI and other advanced technologies. Efforts include hosting panels and events to recruit tech experts across various domains.
AI Challenges and Opportunities: The CIA acknowledges that while AI can be immensely helpful for tasks like data analysis, precision remains a challenge. AI’s output can sometimes be unpredictable, which presents both opportunities and challenges for intelligence operations.
Future Developments: Details about the full scope and capabilities of the CIA’s ChatGPT-style AI are yet to be disclosed, including the specific AI model used as the basis for the project.
The CIA’s investment in AI technology reflects its commitment to staying at the forefront of intelligence gathering and data analysis in an increasingly digital and data-driven world.
OpenAI eyes $90B valuation, dives into AI hardware
OpenAI is in discussions to possibly sell shares, a a move that would boost its valuation from $29 billion to somewhere between $80 billion and $90 billion, according to a Wall Street Journal report citing people familiar with the talks.
In other news, Apple’s former design chief, Jony Ive, and OpenAI CEO, Sam Altman, have reportedly been discussing building a new AI hardware device. It is unclear what the device would be or if they will build it, but the duo has been discussing what new hardware for the AI age could look like.
Why does this matter?
This valuation could provide OpenAI with significant resources to advance AI research and development, perhaps even in hardware. Indeed, if OpenAI explored advancements in AI hardware, it could meet the demand for efficient, specialized hardware to power AI applications that are growing significantly.
Vectara launches Boomerang, the next-gen LLM redefining GenAI accuracy
Outpacing major competitors, Boomerang sets a new benchmark in Grounded Generative AI for business applications. It is a next-generation neural information retrieval model integrated into Vectara’s GenAI platform.
Boomerang surpasses Cohere in benchmark performance and matches OpenAI on certain metrics, excelling particularly in multilingual benchmarks. Notably, it prioritizes security, reducing bias, copyright concerns, and “hallucinations” in AI-generated content. It also offers cross-lingual support for hundreds of languages and dialects and improves prompt understanding, leading to more accurate and faster responses.
Why does this matter?
Vectara’s Boomerang empowers businesses to build AI applications, especially conversational AI, with ease and security, making GenAI more accessible and reliable for enterprise adoption.
Google’s 25-year AI legacy guides its future AI innovations
On its 25th birthday, Google reflected on its two-and-a-half decades of pioneering achievements in the field of AI. It started in 2001 using a simple ML to suggest better spellings for web searches.
A standout moment in 2023 was the introduction of PaLM 2 and Gemini. It is now looking forward to these models driving the next quarter-century of its AI advancements.
Why does this matter?
AI has been a big deal at Google since its earliest days, which predates the emergence of players like OpenAI with ChatGPT. Google’s contributions to AI have had a profound impact not only on its own products and services but have also set industry standards and inspired innovation. Let’s see if it endures its legacy in today’s highly competitive AI landscape.
Why Open Source AI Will Win
As the AI landscape continues to evolve, a crucial future-defining question looms over us: Open or Closed AI?
Varun believes that open source will have more of an impact on the future of LLMs and image models than the broad public believes. In this article, he lists some interesting arguments against open source that he thinks hold little water. He goes on to advocate for open source’s continued growth and adoption.
It also emphasizes their suitability for business-critical applications, their capabilities, control, and customization benefits, as well as their potential for ensuring privacy, security, and reliability.
Why does this matter?
The article challenges the notion that closed-source AI models will dominate the future. It argues that open-source AI models, with their advantages in control, customization, and reliability, will play a significant role in shaping the AI landscape, particularly for business-critical applications.
Microsoft is going nuclear to power its AI ambitions
Microsoft has listed a new job for a “Principal Program Manager Nuclear Technology,” indicating they are exploring Small Modular Reactors (SMRs) for their energy needs.
The new hire will be responsible for integrating SMR and other microreactor designs into Microsoft Cloud’s data centers, creating a roadmap for SMR deployment and identifying cost-saving opportunities.
Despite the current prototype stage of SMRs, Microsoft aims to be at the forefront of using compact nuclear reactors for energy, supplemented by a previous deal with Helion Energy for developing a fusion energy device.
Spotify will not ban AI-made music, says boss
Spotify’s CEO, Daniel Ek, has stated they won’t completely ban AI-created music, though they did remove a song featuring AI-cloned voices of artists Drake and The Weeknd without their consent.
Ek identified three types of AI in music: tools like auto-tune that enhance music, tools that mimic artists (which are not acceptable), and a contentious middle ground where AI-influenced music doesn’t directly impersonate artists.
While AI isn’t entirely banned on Spotify, the platform prohibits its content from being used to train machine learning or AI models that produce music.
Reddit to begin paying people for popular posts
Reddit has launched a new Contributor Program that allows eligible users to earn real-world money from gold they receive on their content.
The program is open to users who are 18 and up, live in an eligible location and meet required karma and gold minimums, with the amount of real-world money earned per gold awarded varying based on these factors.
In line with this, Reddit has made user interface changes, highlighting gold upvotes for awarded content and allowing users to buy and give gold directly from posts, shifting away from Reddit Coins.
Gen Z more susceptible to online scams than boomers
Generation Z, individuals born between the late 1990s and early 2010s, are more likely to fall victim to online scams than older generations, according to a Deloitte survey.
Youths under 20 years old lost an estimated $210 million to online scams in 2022, escalated from $8.2 million in 2017, mainly due to their heavy reliance on the internet and prioritizing convenience over safety.
Vulnerability to “influencer scams” and manipulation by fraudulent websites targeting their online buying habits is a particular risk for Generation Z, who are more at ease with the internet and often overlook the implementation of online security practices.
Daily AI Update News from OpenAI, Vectara, Google, SAP, Microsoft, Shopify and Infosys
OpenAI eyes $90B valuation and dives into AI hardware
OpenAI is in discussions to sell shares, a move that would boost the company’s valuation from $29 billion to somewhere between $80-$90 billion.
Apple’s former design chief, Jony Ive, and OpenAI CEO, Sam Altman have been reportedly discussing building a new AI hardware device. It is unclear what the device would be or if they will build it, but the duo has been discussing what new hardware for the AI age could look like.
Vectara launches Boomerang, the next-gen LLM redefining GenAI accuracy – Outpacing major competitors, Boomerang sets a new benchmark in Grounded Generative AI for business applications, mitigating hallucinations and copyright concerns, minimizing bias, enhancing explainability, and broadening cross-lingual reach.
Google’s 25-year AI legacy guides its future AI innovations – On its 25th birthday, Google reflected on its two-and-a-half decades of pioneering achievements in the field of AI. It started in 2001 using a simple ML to suggest better spellings for web searches. A standout moment in 2023 was the introduction of PaLM 2 and Gemini. It is now looking forward to these models driving the next quarter-century of its AI advancements.
SAP launches its own enterprise AI assistant, Joule – Built into the entirety of SAP’s extensive cloud enterprise suite, Joule will allow customers to access it across SAP apps and programs, similar to Microsoft’s new Windows Copilot. It will also be available across computing platforms, on desktop and mobile.
Microsoft uses AI to boost Windows 11 security, pushes for passwordless future – It announced new enterprise security features that use AI to help defend Windows 11 against increasingly sophisticated cyberattacks. The new AI capabilities may reduce security incidents by 60% and firmware attacks by 300%.
Shopify releases SDXL background replacement tool for product imagery – It is a super helpful tool that can create a whole new reality around your product. Its public HF Space is under the official Shopify account.
Infosys ties with Microsoft for industry-wide adoption of generative AI – The collaboration aims to develop AI solutions, leveraging Infosys Topaz, Azure OpenAI Service, and Azure Cognitive Services. The integrated solutions will enhance enterprise functions and accelerate the democratization of data and intelligence.
Hollywood studios can train AI models on writers’ work under tentative deal – Writers are expected to be guaranteed credit and compensation for work they do on scripts, even if studios partially use AI tools.
A Simple Checklist for Self-Evaluating Prompt Quality
How do you evaluate the quality of your prompt outputs? Here’s a handy checklist. Let’s have a look!
You can also join r/PromptWizards to find more tutorials and prompts!
Part 1: Understanding AI’s Understanding
You’ve presented a prompt to your AI, the next questions are:
Has the AI accurately grasped the context?
If not, how can I make sure the LLM steers my context better, should I be more direct and clear in my prompt? Can I be less negative (shows to perform less) and be more guiding to the LLM?
Do the responses directly address the question or topic?
Was my query and task/instruction clearly detailed in enough depth that the LLM understood what I expect?
Are there any contradictions between different responses to the same prompt?
If I run my prompt multiple times, is the output consistent and reliable?
Are any repetitions apparent in the output, and if so, are they necessary?
Part 2: The Subtleties Matter
The AI’s grasp of finer details can make a world of difference in the generated output. Reflect on these:
Does the language match your output’s expectations?
Were the AI’s responses unbiased?
Did the AI veer off-topic at any stage?
Did the AI ‘hallucinate’ – create any misleading or incorrect information?
Part 3: Deep Evaluation of AI Output
The meaningful evaluation of your AI’s output involves several key areas of consideration:
Was the output’s length and structuring fitting for its intended use?
Did the AI handle nuances, complexities, or subtleties effectively?
Was the AI successful in executing multi-step tasks if they were part of the prompt?
If relevant, were past context or conversations incorporated well into the response?
Could additional guiding examples or context benefit the prompt?
Can the response’s creativity, novelty, or depth be improved?
And finally,
Has the AI displayed a thorough understanding of the user’s set goals?
Did the AI abide by any given constraints in its responses?
Was the AI’s data or factual information accurate and useful?
UNC Researchers Present VideoDirectorGPT: Using AI to Generate Multi-Scene Videos from Text
Generating coherent videos spanning multiple scenes from text descriptions poses unique challenges for AI. While recent progress enables creating short clips, smoothly transitioning across diverse events and maintaining continuity remains difficult.
A new paper from UNC Chapel Hill proposes VIDEODIRECTORGPT, a two-stage framework attempting to address multi-scene video generation:
Here are my highlights from the paper:
Two-stage approach: first a language model generates detailed “video plan”, then a video generation module renders scenes based on the plan
Video plan contains multi-scene descriptions, entities/layouts, backgrounds, consistency groupings – guides downstream video generation
Video generation module called Layout2Vid trained on images, adds spatial layout control and cross-scene consistency to existing text-to-video model
Experiments show improved object layout/control in single-scene videos vs baselines
Multi-scene videos display higher object consistency across scenes compared to baselines
Competitive open-domain video generation performance maintained
The key innovation seems to be using a large language model to plot detailed video plans to guide overall video generation. And the video generator Layout2Vid adds better spatial and temporal control through some clever tweaks. The separation of these tasks seems to matter.
You can read full summary here. There’s a link to the repo there too. Paper link is here.
Colossal-AI has released Colossal-LLaMA-2, an open-source and commercial-free domain-specific language model solution. It uses a relatively small amount of data and training time, resulting in lower costs.
The Chinese version of LLaMA-2 has outperformed competitors in various evaluation benchmarks. The release includes improvements such as vocabulary expansion, a data cleaning system, and a multi-stage pre-training scheme to enhance Chinese and English abilities.
Why does this matter?
This release allows cost-effective training of lightweight domain-specific LLMs, enabling fine-tuning for specific business applications.
The progress made by the open-source community in this field is remarkable, and it raises the question of whether closed models like GPT-4 stand a chance if these open models continue to improve and become more accessible.
Biggest Boom in AI: ChatGPT Talks and Beyond
OpenAI is introducing voice and image capabilities in ChatGPT, allowing users to have voice conversations and show images to ChatGPT. This new feature offers a more intuitive interface and expands the ways in which ChatGPT can be used.
Users can have live conversations about landmarks, get recipe suggestions by showing pictures of their fridge, and even receive math problem hints by sharing photos. The voice and image capabilities will be rolled out to Plus and Enterprise users over the next two weeks, with voice available on iOS and Android and images available on all platforms.
ChatGPT can now comprehend images, including photos, screenshots, and text-containing documents, using its language reasoning abilities. You can also discuss multiple images and utilize their new drawing tool to guide you.
Why does this matter?
OpenAI’s this big feature push comes with ever-rising stakes in the AI race among chatbot leaders such as OpenAI, Microsoft, Google, and Anthropic. These new capabilities to ChatGPT make it a truly multimodal AI and 10x more convenient to use.
It enhances user experiences, expands educational potential, and opens up new horizons in problem-solving. However, they also come with important responsibilities and considerations regarding data privacy and ethical use.
Getty Images’s new AI art tool powered by NVIDIA
Getty Images has launched a generative AI art tool called Generative AI, which uses an AI model provided by Nvidia to render images from text descriptions. The tool is designed to be “commercially safer” than rival solutions, with safeguards to prevent disinformation and copyright infringement.
Getty Images will compensate contributors whose work is used to train the AI generator and share revenues generated from the tool. The tool can be accessed on Getty’s website or integrated into apps and websites through an API, with pricing based on prompt volume. Other companies, including Bria and Shutterstock, are also exploring ethical approaches to generative AI.
Why does this matter?
Getty’s plan to compensate artists and contributors whose work is used to train the AI model highlights the importance of fair compensation and setting a positive example for the industry.
This update enriches user experiences in art, design, and media consumption. They can expect more diverse, high-quality AI-generated content. Using its extensive library responsibly, it aims to create AI content that respects intellectual property rights.
Colossal-AI has released Colossal-LLaMA-2, an open-source and commercial-free domain-specific language model solution. It uses a relatively small amount of data and training time, resulting in lower costs.
The Chinese version of LLaMA-2 has outperformed competitors in various evaluation benchmarks. The release includes improvements such as vocabulary expansion, a data cleaning system, and a multi-stage pre-training scheme to enhance Chinese and English abilities.
Why does this matter?
This release allows cost-effective training of lightweight domain-specific LLMs, enabling fine-tuning for specific business applications.
The progress made by the open-source community in this field is remarkable, and it raises the question of whether closed models like GPT-4 stand a chance if these open models continue to improve and become more accessible.
Spotify makes AI voice clones of podcasters and uses them to speak other languages
Spotify has developed a technology that clones the voices of its top podcasters and uses it to translate their podcasts into other languages.
The voice translation technology is currently available on a limited number of Spanish-language podcasts, with plans to expand it to French and German and include more podcasts.
The AI voice cloning is built on tools provided by OpenAI and was implemented to help build deeper connections and overcome barriers in storytelling, stated by Spotify’s vice president of personalisation, Ziad Sultan.
NASA successfully delivers asteroid samples to Earth
A small capsule with pristine specimens from asteroid Bennu has successfully landed in Utah, marking the end of NASA’s seven-year OSIRIS-REx mission.
The mission has brought back the largest unspoiled sample ever from beyond the Moon, an estimated 250 grams, collected from Bennu during a landing last year.
This is the first asteroid sampling mission for the United States and the third in history, bringing potential insights into the origins of life.
How to write music with ChatGPT: Part 5 – Creating a 90s Rave Hardcore track
The actual tutorial, with the chat transcript and “step-by-step” instructions, can be found here:
Hello, It’s your host again – Low Entropy. Here is another tutorial for creating a track with ChatGPT.
This is a bit different to the other tutorials (well, to be honest, all were different to each other). For example, in part 4, we had a tutorial that was about giving ChatGPT very clear and ordered “instructions”, to get precise answers, essentially to have a full track created by the AI in the end. This is more brain-stormy here, exploring different options, inquiring about things… like a “production dialogue” between me and the AI. When listening to the final track, you will see that while I followed ChatGPT in almost all things in some areas, such as the precise notes and rhythms the AI gave me, the vocals, and so on, in other areas I used the ideas by ChatGPT more as inspiration, improvising and modulating on it, for example when it came to the structure of the track. But I think it’s important to show this side of AI music production, too. That, of course, you don’t always have to follow the AI 100% in each step, but combine it with your own creative efforts, too!
By the way: in order to get the lyrics that ChatGPT created for me into vocals that I could use for the track, I used typecast.ai https://typecast.ai/
Have fun, and enjoy your AI music producing sessions!
Daily AI Update News from OpenAI, NVIDIA, Getty Images, Colossal-AI, Tesla, SnapChat, Microsoft, Spotify, and Google AI
Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the day in the world of AI.
ChatGPT is getting major update, It can now see, hear, and speak – OpenAI is introducing voice and image capabilities in ChatGPT, allowing users to have voice conversations and show images to ChatGPT. – Users can have live conversations about landmarks, get recipe suggestions by showing pictures of their fridge, and even receive math problem hints by sharing a photo. – The voice and image capabilities will be rolled out to Plus and Enterprise users over the next two weeks, with voice available on iOS and Android and images available on all platforms.
Getty Images introduced an art tool called Generative AI, powered by Nvidia – It will render images from text descriptions. The tool is designed to be “commercially safer” than rival solutions, with safeguards in place to prevent disinformation and copyright infringement. – Getty Images will compensate contributors whose works are used to train the AI generator and share revenues generated from the tool. – The tool can be accessed on Getty’s website or integrated into apps and websites through an API, with pricing based on prompt volume.
Colossal-AI released Colossal-LLaMA-2, an open-source and commercial-free domain-specific LLM – It uses a relatively small amount of data and training time, resulting in lower costs. – Includes improvements such as vocabulary expansion, data cleaning system, and a multi-stage pre-training scheme to enhance both Chinese and English abilities. – Allows for cost-effective training of lightweight domain-specific LLMs, enabling fine-tuning for specific business applications.
Tesla’s humanoid robot Optimus can now sort objects autonomously – Using its end-to-end trained neural network. The robot is able to calibrate itself using joint position encoders and vision to locate its limbs precisely. It can then sort colored blocks into their respective trays, even adapting to dynamic changes in the environment. – The robot also uses corrective action to turn blocks right-side-up if they are placed on their side.
Snapchat has partnered with Microsoft to insert ads into its AI chatbot feature, My AI – The chatbot, introduced earlier this year, offers link suggestions related to user conversations. For example, if a user asks for dinner recommendations, the chatbot could reply with a link sponsored by a local restaurant. – The partnership is a win for Microsoft’s ads business and could position Snapchat as a platform for Gen Z users to search for products and services through AI chats.
Spotify is testing a voice translation feature for podcasts, using AI to translate content into different languages – By offering translated podcasts from popular hosts like Dax Shepard and Lex Fridman, Spotify hopes to expand its global reach and cater to a wider audience.
Google’s AI tool, Bard, has now new capabilities to help travelers plan their vacations – By connecting with various Google applications like Gmail, Google Flights, and Google Maps, Bard can provide personalized assistance throughout the trip. – Users can ask Bard to find flight and hotel information, get directions, watch YouTube videos, and even check dates that work for everyone involved. – Additionally, recent updates to Google Flights offer tools to predict the best time to find cheap airline deals.
Correcto has raised $7M in seed funding to expand its language writing tool for Spanish speakers – While AI tools like ChatGPT can generate text in Spanish, Correcto believes its tool offers better quality and provides opportunities for individual learning. The company plans to target enterprise customers while also offering a freemium version for individual users.
Welcome to AI Unraveled, the podcast that demystifies frequently asked questions on artificial intelligence and keeps you up to date with the latest AI trends. Join us as we delve into groundbreaking research, innovative applications, and emerging technologies that are pushing the boundaries of AI. From the latest trends in ChatGPT and the recent merger of Google Brain and DeepMind, to the exciting developments in generative AI, we’ve got you covered with a comprehensive update on the ever-evolving AI landscape. In today’s episode, we’ll cover ChatGPT’s voice capabilities and image inclusion, Amazon’s $4 billion investment in Anthropic, Meta’s plan for various chatbot personas, the efficiency of LongLoRA for extending context sizes of pre-trained LLMs, the differences between RAG and Finetuning LLMs, Coinbase CEO’s opposition to AI regulation, various AI-related news including Meta’s chatbots and Google Pixel 8’s AI camera, and the recommendation to expand AI knowledge with the book ‘AI Unraveled’. Guess what! ChatGPT just got an awesome upgrade! Now, it’s not just about typing and reading messages. It can actually see, hear, and even talk! How cool is that? Hold on, there’s more! If you’re a Plus user, you’re going to love this update. Over the next two weeks, you’ll be able to have voice conversations with ChatGPT, whether you’re using iOS or Android. Yep, you heard that right. You can actually have conversations with ChatGPT using your own voice! It’s like having a real back-and-forth chat with a super smart AI buddy. But wait, there’s one more exciting thing coming your way. Are you tired of explaining things only with words? Well, now you don’t have to! With this update, all platforms will allow you to include images in your conversations. That means you can now send pictures to ChatGPT to help illustrate what you’re talking about. It’s another way to make your conversations more engaging and dynamic. So, strap in and get ready for an even more immersive experience with ChatGPT. It’s no longer just a text-based AI companion—it’s now a full-on interactive conversational partner that can see, hear, and speak. Enjoy the future of chat! So, get this: Amazon is planning to invest a whopping $4 billion in Anthropic. But wait, what’s Anthropic, you ask? Well, it’s this company that’s all about developing the most reliable and high-performing foundation models in the industry. Here’s the game-changing part: Anthropic’s safety research and products, along with the expertise of Amazon Web Services (AWS) in running secure and reliable infrastructure, will make Anthropic’s safe and controllable artificial intelligence (AI) accessible to AWS customers. In other words, they’re joining forces to bring us safe and steerable AI in a big way. Now, here’s where things get even more interesting. AWS is set to become Anthropic’s primary cloud provider for those mission-critical workloads. And that’s not all—they’re also expanding Anthropic’s support of Amazon Bedrock, whatever that may be. But why is this such a big deal, you ask? Well, this collaboration will open up new possibilities for enterprises. It means they can build with Anthropic models on Amazon Bedrock and responsibly scale the adoption of Claude (no, not the guy down the street, but another AI model). This could revolutionize the delivery of safe AI cloud technologies to organizations worldwide. So, brace yourselves, folks. We’re about to witness some serious advancements in the world of AI, courtesy of Amazon and Anthropic! So, here’s a juicy update! Meta, the tech giant, has got some interesting plans in the pipeline. Apparently, they’re working on creating a bunch of sassy chatbot ‘personas’ specifically designed to captivate and interact with younger users. I mean, who wouldn’t want a chatbot with a colorful personality, right? But wait, there’s more! Meta isn’t stopping there. They’re also cooking up some chatbot personas for celebrities to connect with their devoted fans. Imagine having a virtual conversation with your favorite superstar! And if that wasn’t enough, they’ve got some chatbots that are all about productivity too. These ones are geared towards helping with coding and other nifty tasks. Now, why is all of this such a big deal? Well, rumor has it that Meta’s got something else up their sleeve. They’re apparently working on a super-powered LLM (that’s a language model, by the way) to rival the likes of OpenAI. This could be a major leap forward in AI capabilities, my friends. And hey, it might also give a boost to engagement on Meta’s social media platforms. Exciting times ahead, folks! Meta’s bringing some sass and brains to the chatbot game. Keep your eyes peeled for these new chatty personalities hitting your screens soon! Today, we’re talking about a new research development called LongLoRA. This method aims to make fine-tuning of long-context Language Models (LLMs) more efficient, without requiring a lot of computational power. The ability to extend the context size of pre-trained LLMs is crucial in many natural language processing tasks. Traditionally, training LLMs with longer context sizes comes with a hefty computational cost and demands powerful GPU resources. For example, increasing the context length from 2048 to 8192 can lead to a 16-fold increase in computational costs, especially in self-attention layers. However, LongLoRA tackles this challenge by employing two main strategies. First, it utilizes sparse local attention instead of dense global attention, which can be optional during inference. This helps in reducing the computational burden. Second, LongLoRA incorporates LoRA (Low-Rank Adaptation) for context extension. The beauty of LongLoRA lies in its simplicity. It has shown promising results across various tasks using LLaMA-2 models, ranging from 7B/13B to 70B. Notably, it successfully extended the context size of LLaMA-2 7B from 4k to 100k and LLaMA-2 70B to 32k using just a single 8x A100 machine. What’s impressive is that these extensions were achieved while keeping the original model architectures intact. So, why does this matter? Well, LongLoRA is a significant step forward in creating more computationally efficient model expansion. If you’re interested in developing open-source LLMs with longer context lengths, LongLoRA might just be the solution that lowers the barrier to entry. In the world of language models, there are two popular methods that many AI developers use with “custom” data: RAG (Retrieval Augmented Generation) and finetuning. But here’s the thing – it can be confusing to determine which method to use, when to use it, and why. Luckily, John Hwang, in his insightful article, dives deep into this topic to clarify everything for us. First and foremost, Hwang points out that RAG and finetuning are fundamentally different tools meant for different problems. To help us understand better, he even includes a table comparing the two methods. Not stopping there, Hwang also lists out the right use cases for RAG and finetuning. He explains that these methods have their own strengths and weaknesses, and it’s important to consider the specific problem you’re trying to solve. But wait, there’s more! Hwang also provides us with a list of other factors we should consider when contemplating RAG and finetuning. These factors could range from the availability of training data to the constraints of real-time applications. To wrap it all up, Hwang presents a set of heuristics – essentially guidelines – for choosing the appropriate method depending on the situation. This is immensely helpful for AI developers who often find themselves stuck in analysis paralysis or caught up in premature optimization. Ultimately, this article matters because it not only helps AI developers navigate the complexities of RAG and finetuning, but it also supports enterprises in making well-informed investment decisions. By clarifying when and how to apply these methods effectively, Hwang ensures that we don’t waste our time, effort, and resources on the wrong approach. Hey everyone! I’ve got some interesting news for you today. Coinbase CEO, Brian Armstrong, recently expressed his opposition to regulating artificial intelligence (AI). He believes that imposing regulations on AI could actually hinder innovation. Armstrong is all about a “decentralize” and “open source” approach when it comes to AI. If you’re familiar with his stance on the crypto industry, this might sound familiar to you. He sees similar potential in AI and believes that a more flexible approach is needed. According to Armstrong, rapid progress in AI is of utmost importance. He even cites national security as one of the reasons why we need to keep pushing forward. It’s true that AI has its dark side, with nefarious uses like promoting crypto scams, but Armstrong emphasizes the benefits outweigh the risks. In a world that’s constantly evolving, we need innovation to thrive. And that means embracing new technologies like AI. So, according to Armstrong, let’s keep the regulations at bay and allow AI to continue its rapid development. It’s an interesting perspective, and it’ll be fascinating to see how this debate unfolds. Hey there! I’ve got some interesting AI updates for you today. Let’s jump right in! First up, Amazon is investing a whopping $4 billion in Anthropic. This collaboration aims to develop the most reliable and high-performing foundation models. Anthropic is known for its frontier safety research and products, while AWS brings its expertise in running secure and reliable infrastructure. Together, they will make Anthropic’s safe and steerable AI widely accessible to all AWS customers. Exciting stuff! Moving on, Meta has some cool plans for their AI chatbot. They want to develop a range of chatbot personas that cater to different users. There will be personas for engaging younger users with more colorful behavior, as well as ones for celebrities to interact with their fans. They also have chatbots geared towards productivity, like helping with coding and other tasks. Meta is really bringing chatbots to life! Now, let’s talk about some new research called LongLoRA. This method allows for efficient fine-tuning of long-context Language Models (LLMs). The goal here is to extend the context sizes of pre-trained LLMs without a huge computation cost. LongLoRA has shown impressive performance on various tasks using LLaMA-2 models, ranging from 7B/13B to 70B. It can extend the context size of LLaMA-2 7B from 4k to 100k and LLaMA-2 70B to 32k on a single 8x A100 machine. And the best part? It keeps the original model architectures intact. Quite impressive! Next up, Microsoft’s mobile keyboard app SwiftKey is getting some AI-powered features. Get ready for AI camera lenses, AI stickers, an AI-powered editor, and the ability to create AI images directly from the app. Now you can take your mobile photography game to the next level with these cool additions. SwiftKey is really stepping up its game! Speaking of AI camera updates, the latest leak about Google Pixel 8 has got us excited. The AI photo editing with Magic Editor is going to blow your mind. You’ll be able to remake any picture you take using this feature. And that’s not all! The Pixel 8 will also have DSLR-style manual camera controls, allowing you to tweak the shutter speed and ISO of an image. Plus, there will be a focus slider for that perfect shot. Get ready to capture some stunning photos with the Google Pixel 8! Now here’s something unique. A drinks company in Poland, Dictador, has appointed an AI robot as its “experimental” CEO. This robot, named Mika, will oversee the company’s growth into one-off collectables, communication, and even strategy planning. It’s definitely an interesting move to have an AI robot at the helm of a company. We’ll have to wait and see how this experiment unfolds! If you’re a fan of classic stories, you’re in for a treat. ElevenLabs has launched free book classics narrated by high-quality AI voices. Now you can enjoy stories like “Winnie the Pooh” and “The Picture of Dorian Gray” narrated by compelling AI voices in multiple languages. And guess what? The entire recording process took only one day. Talk about efficiency! Last but not least, Salesforce is making moves in the AI space. They’re set to acquire Airkit.ai, a low-code platform for building AI customer service agents. This platform, based on GPT-4, allows e-commerce companies to build specialized customer service chatbots. These chatbots can handle queries related to order status, refunds, product information, and more. It’s all about enhancing the customer service experience with the power of AI. And that’s a wrap for today’s AI update news! From investments in safer AI to AI-powered features in various apps, the world of AI continues to evolve and amaze us. Stay tuned for more exciting updates in the future! Hey there! If you’re excited about diving deeper into the world of artificial intelligence, I’ve got just the thing for you! There’s this amazing book called “AI Unraveled: Demystifying Frequently Asked Questions on Artificial Intelligence.” Trust me, it’s a game-changer! Now, let me tell you why you should totally get your hands on this gem. “AI Unraveled” is packed with all the answers to those burning questions you may have about AI. Think of it as your ultimate AI guidebook. It’s like having a knowledgeable expert right by your side, unravelling the mysteries of artificial intelligence in a way that’s easy to comprehend. The best part? You can grab a copy of this must-read book at three different platforms: Apple, Google, or Amazon. So, no matter whether you’re an Apple aficionado, a Google guru, or an Amazon enthusiast, there’s a way for you to access this invaluable resource. So, why wait any longer? Dive into “AI Unraveled” today and expand your understanding of artificial intelligence like never before. This book is a game-changer, and it’s ready to be enjoyed by curious minds like yours. Happy reading! On today’s episode, we covered the addition of voice capabilities and image inclusion in ChatGPT, Amazon’s $4 billion investment in Anthropic for reliable AI models, Meta’s plan to create various chatbot personas, the efficient method of LongLoRA for extending context sizes of LLMs, insights on RAG vs. Finetuning LLMs, Coinbase CEO’s opposition to AI regulation, and other AI updates including Google Pixel 8’s AI camera and Salesforce’s acquisition of Airkit.ai for AI customer service agents. Don’t forget to expand your AI knowledge with the essential book ‘AI Unraveled’ available at Apple, Google, or Amazon! Join us next time on AI Unraveled as we continue to demystify frequently asked questions on artificial intelligence and bring you the latest trends in AI, including ChatGPT advancements and the exciting collaboration between Google Brain and DeepMind. Stay informed, stay curious, and don’t forget to subscribe for more!
—–
Are you eager to expand your understanding of artificial intelligence? Look no further than the essential book “AI Unraveled: Demystifying Frequently Asked Questions on Artificial Intelligence,” available at Apple, Google, or Amazon today!
Rolling out over next two weeks, Plus users will be able to have voice conversations with ChatGPT (iOS & Android) and to include images in conversations (all platforms).
Amazon to Invest $4B in Anthropic
Amazon will invest up to $4 billion in Anthropic. The agreement is part of a broader collaboration to develop the industry’s most reliable and high-performing foundation models.
Anthropic’s frontier safety research and products, together with Amazon Web Services’ (AWS) expertise in running secure, reliable infrastructure, will make Anthropic’s safe and steerable AI widely accessible to AWS customers. AWS will become Anthropic’s primary cloud provider for mission-critical workloads, and this will also expand Anthropic’s support of Amazon Bedrock.
Why does this matter?
It will enable enterprises to build with Anthropic models on Amazon Bedrock, responsibly scaling the adoption of Claude and delivering safe AI cloud technologies to organizations worldwide.
Meta to develop a ‘sassy chatbot’ for younger users
Meta has plans to develop dozens of chatbot ‘personas’ geared toward engaging young users with more colorful behavior. It also includes ones for celebrities to interact with their fans and some more geared towards productivity, such as to help with coding and other tasks.
Why does this matter?
Reportedly, Meta is also working on developing a more powerful LLM to rival OpenAI. Perhaps this could serve as a stepping stone towards more advanced AI capabilities and also boost engagement on Meta’s social media platforms.
LongLoRA: Efficient fine-tuning of long-context LLMs
New research has introduced LongLoRA, an ultra-efficient fine-tuning method designed to extend the context sizes of pre-trained LLMs without a huge computation cost.
Typically, training LLMs with longer context sizes consumes a lot of time and requires strong GPU resources. For example, extending the context length from 2048 to 8192 increases computational costs 16 times, particularly in self-attention layers. LongLoRA makes it way cheaper by:
1. Using sparse local attention instead of dense global attention (optional at inference time).
2. Using LoRA (Low-Rank Adaptation) for context extension
This approach seems both easy to use and super practical. LongLoRA performed strongly on various tasks using LLaMA-2 models ranging from 7B/13B to 70B. Notably, it extended LLaMA-2 7B from 4k context to 100k and LLaMA-2 70B to 32k on a single 8x A100 machine, all while keeping the original model architectures intact.
Why does this matter?
LongLoRA is an important step toward making model expansion more computationally efficient. For those interested in creating open-source LLMs with longer context lengths, LongLoRA may be the lowest barrier to entry.
RAG vs. Finetuning LLMs – What to use, when, and why
RAG (Retrieval Augmented Generation) and finetuning are two popular methods for using LLMs with “custom” data. However, it can be confusing to know which method to use, when, and why.
Clarifies that RAG and finetuning are fundamentally different tools for different problems. (includes a table comparing the two)
Lists out the right use cases of RAG and finetuning.
Lists out other factors to consider when considering RAG and finetuning.
Presents a set of heuristics for choosing what method to use and when.
Why does this matter?
The article helps AI developers navigate between the two methods and avoid analysis paralysis and premature optimization. Moreover, it assists enterprises in making informed investment decisions by clarifying when and how to apply these methods effectively.
Coinbase CEO calls for AI deregulation
Coinbase CEO Brian Armstrong opposes regulation on artificial intelligence, arguing it would slow innovation.
Armstrong supports a “decentralize” and “open source” approach in AI, similar to his stance on the crypto industry.
Fast progress on AI, he argues, is critical for various reasons including national security, and despite nefarious uses like promoting crypto scams.
Amazon to invest up to $4 billion in Anthropic, expanding access to safer AI – It is part of a broader collaboration to develop the most reliable and high-performing foundation models. Anthropic’s frontier safety research and products, together with AWS’s expertise in running secure, reliable infrastructure, will make Anthropic’s safe and steerable AI widely accessible to AWS customers.
Meta’s AI chatbot plan includes a ‘sassy robot’ for younger users – Meta has plans to develop dozens of chatbot personas geared towards engaging young users with more colorful behavior. It also includes ones for celebrities to interact with their fans and some more geared towards productivity, such as to help with coding and other tasks.
LongLoRA: Efficient fine-tuning of long-context LLMs – New research has introduced LongLoRA, an efficient fine-tuning method designed to extend the context sizes of pre-trained LLMs without a huge computation cost. In practical terms, LongLoRA performed strongly on various tasks using LLaMA-2 models ranging from 7B/13B to 70B. Notably, it extended LLaMA-2 7B from 4k context to 100k and LLaMA-2 70B to 32k on a single 8x A100 machine, all while keeping the original model architectures intact.
Microsoft’s mobile keyboard app SwiftKey gains new AI-powered features – It will now include AI camera lenses, AI stickers, an AI-powered editor, and the ability to create AI images from the app.
Google Pixel 8’s latest leak shows off big AI camera updates – AI photo editing with Magic Editor will enable you to remake any picture you take. DSLR-style manual camera controls will let you tweak the shutter speed and ISO of an image and a focus slider.
A drinks company in Poland appoints AI robot as ‘experimental’ CEO – Dictador, best known for its rums, has appointed the robot to oversee the company’s growth into one-off collectables, communication, or even strategy planning. It is named Mika.
ElevenLabs launches free book classics narrated by high-quality AI voices – It presents six classic stories told by compelling AI voices in multiple languages, including “Winnie the Pooh” and “The Picture of Dorian Gray.” The entire recording process took only one day.
Salesforce to acquire Airkit.ai, a low-code platform for building AI customer service agents – The GPT-4-based platform allows e-commerce companies to build specialized customer service chatbots that can deal with queries around order status, refunds, product information, and more.
Emerging AI Innovations: September 24th, 2023
Researchers discover ‘Reversal Curse:’ LLMs trained on “A is B” fail to learn “B is A”
Training AI models like GPT-3 on “A is B” statements fails to let them deduce “B is A” without further training, exhibiting a flaw in generalization. (Full 18 pg. paper)
The Phenomenon
Models can’t infer “B is A” after learning “A is B.”
E.g. won’t deduce “Olaf Scholz was the ninth Chancellor” from opposite statement.
True even for models with billions of parameters.
Evidence of Deficiency
Tested models on fictitious “X is Y” statements and reverse questions.
Evaluated ChatGPT on real celebrity examples and their reversals.
Success rate dropped from 79% to 33% between forward and reverse.
TL;DR: Research exposed the “reversal curse” showing models can’t infer reversed statements, underscoring flaws in logical generalization capabilities.
Artificial Intelligence Could Finally Let Us Talk with Animals
So, imagine this: a New Caledonian Crow meticulously crafting a tool to fish out a tasty grub from a tree crevice. Fascinating, right? Turns out, animals have some incredible communication skills that we’ve only just begun to understand. Take Christian Rutz, a behavioral ecologist who has spent his career studying the New Caledonian Crow. He noticed that these birds live in complex social groups and actually pass on toolmaking techniques to their offspring. But here’s the kicker – different crow groups have their own unique vocalizations. Rutz wondered if these dialects could explain the cultural differences in toolmaking among the groups. That’s where artificial intelligence (AI) comes in. With recent advancements, we’re on the brink of major breakthroughs in understanding animal communication. AI can help us decipher animal vocalizations, like the calls of crows. A group called the Earth Species Project is using machine learning models to analyze data collected from various species. And the Project Cetacean Translation Initiative is specifically focused on understanding the vocalizations of sperm whales. Decoding these vocalizations not only aids conservation and welfare efforts, but it also has a profound impact on us. It’s like when the telescope was invented – we realized Earth wasn’t the center of the universe. AI has the potential to reshape our understanding of animals and our place in the world. Scientists like Shane Gero are already using AI to decode the complex vocalizations of sperm whales. By analyzing patterns of sound, called codas, Gero and his team have identified individual whales with 99 percent accuracy. And now they’re aiming to train a computer to speak whale by recording the vocalizations of Dominica’s resident whales around the clock. AI is opening up a whole new world of possibilities for understanding animal communication, and it’s truly mind-blowing. These tools are changing the way we see ourselves in relation to everything around us.
Artificial intelligence (AI) has made significant progress in recent years, enabling us to delve into the world of animal communication like never before. With the availability of cheaper sensors and advancements in technology such as hydrophones, biologgers, and drones, the amount of data gathered from animals has exploded. However, this deluge of data is challenging for biologists to manually analyze efficiently. This is where AI comes in. AI models, particularly large language models like ChatGPT, thrive on vast amounts of information. For instance, ChatGPT-3 was trained on approximately 45 terabytes of text data, a substantial portion of the entire Library of Congress. In the early days, humans had to classify a significant part of this data with labels, essentially teaching the machines what was important. Nevertheless, the next generation of models have advanced to the point where they can “self-supervise.” They automatically learn the essential elements and create algorithms to predict what words will follow in a sequence. A breakthrough in translation occurred in 2017 when researchers found a way to translate between human languages without relying on a Rosetta stone. This discovery involved transforming the semantic relations between words into geometric ones. Machine-learning models can now translate unknown human languages by aligning shapes. By analyzing the frequency of words appearing in proximity to each other, models accurately predict what will follow. This suggests that there is an underlying structure that unites languages, opening doors for decoding new languages using machine learning techniques. In 2020, natural-language processing took another step forward by considering everything as a language. For example, DALL-E 2, an AI system capable of generating realistic images from verbal descriptions, maps the shapes that represent text to those that represent images with remarkable accuracy. This kind of “multimodal” analysis is likely to be crucial in translating animal communication, as many animals use different modes of communication concurrently, much like humans use body language alongside speech. By considering the actions immediately before, during, or after sounds, we can better understand the context and meaning behind an animal’s communication. Machine-learning models trained with the right data could help decode these behaviors and potentially uncover new patterns. An example of AI-powered analysis already in use is Merlin, a free app developed by the Cornell Lab of Ornithology. Merlin uses AI to identify bird species. Users can record bird sounds, which Merlin then converts into a spectrogram, a visual representation of the volume, pitch, and duration of the call. By comparing the user’s recording with its trained audio library and cross-referencing with the global database of observations called eBird, Merlin can accurately identify more than 1,000 bird species. This demonstrates the potential for AI to assist in understanding animal communication on a broader scale. The ability of AI to process and analyze large amounts of data can greatly aid in deciphering complex animal communication patterns. In a study published in Nature Communications, scientists reported that machine-learning models unearthed previously unrecognized differences in Zebra Finch songs that elicit attention from females during mate selection. Females prefer partners that sing similar to the birds they grew up with, and AI models were able to identify this pattern. Such findings highlight the potential for AI to discover novel information hidden within animal communication data. AI’s impact on animal communication research goes beyond language translation and species identification. It has the potential to detect and understand the nuances of animal behavior by examining the context in which communication occurs. For instance, AI models trained to recognize patterns in body language and gestures, along with vocalizations, could reveal valuable insights into how animals communicate and interact with each other. With the rapid advancements in AI and the increasing availability of data, the future of animal communication research looks promising. By harnessing the power of AI, scientists can explore, decode, and understand the rich world of animal communication in ways that were previously unimaginable. It is an exciting frontier that opens up a new realm of knowledge and understanding about our fellow creatures with whom we share this planet.
But you know, the world can be a noisy place, with so many sounds overlapping and mingling together. It’s like trying to pick out one specific bird’s song in the middle of a bustling cacophony. And the same goes for other animals, like whales. It has always been a challenge for scientists to isolate and identify individual animal speakers amidst all the chaos. They call it the cocktail party problem, and it has always hindered our ability to process and understand animal vocalizations. But here’s the exciting part – in 2021, the Earth Species Project came up with a groundbreaking solution. They developed a neural network that can untangle and separate overlapping animal sounds, like individual tracks in a music recording. And you know what’s even better? They released the code for free, as an open-source project. This means that anyone can use this technology to filter out background noise, like car honks, and focus on the specific sounds they want to study. It’s a game-changer! Let me explain how it works. This neural network creates a visual representation of the animal sounds, like a spectrogram or a waveform. It then uses this representation to determine which pixel corresponds to which speaker. It’s like a magic trick for audio processing. Now researchers can finally distinguish between different animals in a chorus of sounds and understand each speaker individually. And that’s not all! The Earth Species Project has been on fire lately. They also developed what they call a foundational model. This model can automatically detect and classify patterns in massive datasets. Imagine the possibilities! It can help scientists analyze and make sense of the immense amount of information contained in animal vocalizations. It’s like a revolutionary tool for unlocking the secrets hidden in the animal kingdom’s communication systems. Let me give you an example. Have you heard of the New Caledonian Crows? They are renowned for their remarkable tool-making skills. But did you know that they also have unique vocalizations specific to different regions? It’s fascinating! And with the help of AI, we might one day decipher and understand the meaning of these vocalizations. Just imagine the insights we could gain into the lives of these intelligent creatures. But it’s not just about unraveling the mysteries of nature. These tools have real practical value as well. Think about endangered species, like the Hawaiian Crow, or the ‘Alalā as it is known locally. These birds went extinct in the wild a few decades ago, and efforts have been made to conserve and reintroduce them through breeding programs. One of the researchers involved in studying the New Caledonian Crows, Rutz, is now collaborating with the Earth Species Project to delve into the Hawaiian Crow’s vocalizations. He wants to create an inventory of the vocalizations used by the captive birds and compare them to historical recordings of the last wild Hawaiian Crows. By doing so, he hopes to uncover whether there have been any significant changes in their repertoire while in captivity. This knowledge could aid in understanding why reintroducing the crow to its natural habitat has been so challenging. Now let’s take a moment to talk about our beloved pets. For the longest time, animal behaviorists haven’t paid much attention to domestic pets. But that is changing. Con Slobodchikoff, the author of “Chasing Doctor Dolittle: Learning the Language of Animals,” has been studying prairie dogs for years and has discovered the complexity of their communication through calls. This understanding led him to become a behavioral consultant for dogs, as he realized that many owners misinterpret their pets’ signals. You see, our furry friends don’t just rely on barks; they communicate through various signals, including body language. And unfortunately, we often miss out on these cues because we are fixated on sound as the primary means of communication. But Slobodchikoff is working on an AI model that can help translate a dog’s facial expressions and barks for their owners. He firmly believes that animals have their own thoughts, hopes, and even dreams. And with the help of AI, we might just be able to understand them better and strengthen the bond between humans and our four-legged companions. But the potential impact of this technology extends beyond companion animals. Farm animals, for instance, could greatly benefit from a deeper understanding of their emotions. Elodie F. Briefer, an associate professor in animal behavior, has been studying pig vocalizations and their emotional states. She trained an algorithm using thousands of pig sounds to predict whether the animals were experiencing positive or negative emotions. This incredible feat demonstrates the potential for AI and machine learning to improve animal welfare by helping us comprehend their feelings and tailor better care. So you see, these developments in AI and machine learning aren’t just for the sake of research. They have practical applications that can make a difference in the world. Whether it’s protecting endangered species, understanding our pets better, or ensuring the welfare of farm animals, these tools offer us a glimpse into the rich and complex world of animal communication. And who knows what other surprises await us as researchers continue to explore and unlock the secrets of the animal kingdom?
Language models are incredibly proficient at identifying patterns, but they do not possess the ability to decipher meaning or always make accurate conclusions. This lack of understanding makes it difficult for AI experts to validate the algorithms’ results. Benjamin Hoffman, formerly involved in the development of the Merlin app and currently a member of the Earth Species Project, states that one of the biggest challenges scientists face is how to learn from the discoveries made by these models. Hoffman explains that the choices made in machine learning have a direct impact on the scientific questions that can be asked. For example, Merlin Sound ID can identify which birds are present, aiding in ecological research. However, it cannot answer questions about bird behavior, such as the types of calls made during interactions with potential mates. Understanding animal communication requires comprehending what the computer is doing when learning how to interpret it. Director Daniela Rus from the Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory is excited about the possibilities of studying animal communication with the help of machine learning. Rus previously developed remote-controlled robots for whale-behavior research, collaborating with biologist Roger Payne, known for popularizing the Save the Whales movement. With advancements in underwater monitoring sensors and improved AI models for data analysis, the two fields can now be combined. At the Earth Species Project, Rus initially focused on isolating sperm whale clicks from the ocean’s background noise. Sperm whale vocalizations resemble binary code in the structure of representing information, but they are more intricate than that. Rus used machine learning to analyze how these clicks combine into codas, searching for patterns and sequences. By understanding these building blocks, they can begin studying the foundational components of the language and determining if the sperm whale lexicon possesses language-like properties. It is worth noting that understanding the structure of a language is not necessary for speaking it. AI can now mimic patterns and intonations of human speech after being trained with just three seconds of audio. Raskin, a project member, predicts that within the next year or two, this capability will extend to animal communication, making it easier for researchers to infer the meaning behind animal vocalizations. The Earth Species Project plans to conduct playback experiments in collaboration with biologists, involving playing artificially generated calls to animals in a laboratory setting and observing their responses. Raskin confidently asserts that soon they will be able to pass the Turing test with animals such as Zebra Finches, crows, and whales. This test determines if an animal can differentiate between conversations with a machine or a member of its own species. However, ethical concerns arise with the potential misuse of this technology, such as precision fishing or poaching endangered animals. As of now, non-profit organizations like the Earth Species Project lead the field of animal communication research. These organizations prioritize open-source data and model sharing, driven by scientists passionate about the animals they study. However, the landscape may change as profit-driven entities enter the scene. Guidelines and legislative frameworks are necessary to ensure responsible development and usage of this technology. Designing a “whale chatbot,” like the aspirations of Project CETI, entails more than replicating the clicks and whistles of sperm whales—it requires understanding the animal’s experience. Humans share many basic forms of communication with other animals, such as parent-offspring interactions. Both human and animal vocal expressions show similarities in development. The existence and conveyance of language-like qualities in animal communication are a subject of debate, with critics cautioning against imposing human linguistic rules on it. George Happ and Christy Yuncker, retired scientists who observed a pair of wild Sandhill Cranes over two decades, offer insight into the behavior of these birds. They explain how after the death of one of their colts, the surviving members engaged in what could be interpreted as mourning behavior. While critics may argue that squarely attributing emotions to animals lacking human capabilities is imprecise, the evidence from close observation suggests otherwise. The pain of losing a loved one is a universal experience, and the ability to relate to it may be the true value of any language. Yuncker and Happ eagerly anticipated the return of Millie and Roy, a crane pair that frequented their home every spring. Unfortunately, they did not appear in 2017. Despite the loss, a new crane pair nested and successfully raised their colts, signifying the continuation of life’s cycles. Yuncker remarks that, rather than just observing nature, humans are an integral part of it. The understanding and translation of animal communication necessitate new tools and the ability to transcend human biases and expectations. Recognizing that each species has its own unique experiences and perspectives is vital for comprehensive communication with animals.
In recent years, the ML community has focused on training increasingly large and powerful self-supervised (language) models. Since these LLMs exhibit impressive predictive capabilities, they are well-positioned to be strong compressors.
This interesting research by Google DeepMind and Meta evaluates the compression capabilities of LLMs. It investigates how and why compression and prediction are equivalent. It shows that foundation models, trained primarily on text, are general-purpose compressors due to their in-context learning abilities. For example, Chinchilla 70B achieves compression rates of 43.4% on ImageNet patches and 16.4% on LibriSpeech samples, beating domain-specific compressors like PNG (58.5%) or FLAC (30.3%), respectively.
Emerging AI Innovations: September 22nd, 2023
Microsoft recently announced a game-changing feature called Microsoft Copilot. This exciting new addition will infuse AI capabilities into various Windows 11, Microsoft 365, Edge, and Bing applications. Think of it as Bing, but specifically designed for Windows devices. So, what can Copilot do? Quite a lot, actually. With this tool, you can rearrange windows effortlessly, generate text, open web apps, edit pictures, and much more. It’s accessible both via an app and through a simple right-click, making it convenient for users to tap into its AI-powered goodness. But when can we start using Copilot? Well, the good news is that it’s just around the corner. Microsoft plans to roll out Copilot this fall, making it available across Bing, Edge, and Microsoft 365. And for Windows users, you’ll get to enjoy this feature sooner than you think. The free Windows 11 update will begin on September 26th. Now, you might be wondering, why is this such a big deal? The answer lies in the democratization of AI. While we don’t have any mind-blowing use cases for Copilot just yet, this step forward by Microsoft is significant. As more users get their hands on this AI copilot, we’ll start to see its true capabilities. And if all goes well, Microsoft could dominate an even larger share of the AI market by delivering AI nativel
Hey folks! YouTube just announced some exciting news for creators! They’re rolling out three new AI-powered features for YouTube Shorts creators. Let me break it down for you. First up, we have Dream Screen. This feature lets you create image or video backgrounds using AI. All you have to do is type in what you want to see in the background, and AI will make it happen. How cool is that? Next, we’ve got Creator Music. This feature got an AI revamp, making it even better than before. Now, creators can simply type in the kind and length of the music they need, and AI will find the most relevant suggestions. It’s like having your own personal music assistant. Last but not least, we have AI Insights for Creators. This is a tool that generates video ideas for creators based on AI’s analysis of what audiences are already watching and preferring. So, if you’re looking for some inspiration, AI has got your back. This move by YouTube seems like a smart strategic decision to integrate AI features directly into the platform. We’re seeing this trend more and more, and it’s great news for users. They get free AI assistance in their creative endeavors, making their experience even better. In other news, Google is taking an innovative approach to train smaller language models. You know those large language models (LLMs) that have been making waves? Well, their massive size poses some deployment challenges. But fear not! The authors propose a method called distilling step-by-step, which trains smaller task-specific models using less data while still surpassing LLM performance. Here’s the gist: they extract rationales (aka intermediate reasoning steps) from an LLM using few-shot chain-of-thought prompting. These rationales, along with labels, are then used to train smaller models in a multi-task framework. The results? In experiments across different datasets, this approach reduced the need for training data by a whopping 75-80% compared to standard fine-tuning. Why does this matter? Well, this new approach opens up possibilities for deploying language models on local devices, making them more accessible. And the best part? These smaller models can still deliver the performance we’ve come to expect from their larger counterparts. That’s all for now, folks! Exciting times ahead for creators and language models alike. Keep creating and exploring!
When it comes to evaluating large language models (LLMs) for industry applications, there are four crucial factors to consider. Skanda Vivek highlights these factors, which include quality, economic aspects, latency, and privacy. Each of these factors plays a significant role in determining the suitability of a particular LLM. The quality of the LLM is of utmost importance. Depending on your end goal, you may prioritize different aspects of quality, such as data accuracy, contextual understanding, or fluency. Consider what matters most to your industry and choose an LLM that aligns with those preferences. Economic factors also come into play. It’s essential to assess the cost-effectiveness of implementing a particular LLM. Does it provide value for money? Can it fit within your organization’s budget? Analyzing the economic aspects ensures you make an informed decision. Latency, or the response time of the LLM, is another vital factor. Some applications require real-time or near-instantaneous responses. Evaluating an LLM’s latency helps you select the model that meets your specific timing requirements. Finally, privacy is increasingly significant for many industries. Skanda Vivek emphasizes the need to consider privacy when choosing an LLM. Depending on your industry, data security and privacy regulations may be a top priority. Ensuring the chosen model aligns with your privacy needs is crucial. Choosing the right LLM is a critical decision that can significantly impact your applications. By carefully considering these four factors—quality, economic aspects, latency, and privacy—you can make an informed choice that aligns with your industry’s requirements. In recent news, some universities are raising concerns about AI detection software used to catch cheating students. There are worries that students could be falsely accused of cheating when using tools like ChatGPT. As a result, some universities are opting to abandon these AI detection systems. The debate highlights the potential drawbacks and risks associated with relying entirely on AI tools for academic integrity.
So, here’s the thing. Some major universities have decided to ditch AI detection tools because they’re worried about their accuracy. And let’s face it, nobody wants to be falsely accused of cheating, right? One tool in particular, called ChatGPT, has caused quite a stir. The problem with ChatGPT is that it’s gained popularity among students, and that’s got educators really concerned about academic dishonesty. But it’s not just about students using AI to write their essays. It’s also about the tool itself misidentifying things and getting it all wrong. For example, one professor in Texas failed half of his class because of false detections by ChatGPT. Can you imagine? Talk about a nightmare scenario. And it’s not just him. Other students have also been wrongly accused by anti-plagiarism software using ChatGPT. What’s interesting is that even OpenAI, the company behind ChatGPT, has abandoned their own AI text detector due to its low accuracy rate. They’ve even warned educators about relying too heavily on AI content detectors. And here’s another thing to consider: these detection tools often get it wrong when it comes to content written by non-English writers. So, yeah, there are some serious concerns here. That’s why some universities, like Vanderbilt and Northwestern, have decided to say “no thanks” to these AI detection tools. It’s better to be safe than sorry, right? After all, nobody wants to unfairly accuse a student of cheating.
Hey there! Some interesting news for you today. According to Climate Action Against Disinformation, X, which we all know as Twitter, has ranked last when it comes to tackling climate misinformation. Quite the bummer, right? It turns out that Pinterest is leading the pack in addressing climate change misinformation, with YouTube, Meta (formerly known as Facebook), and Instagram not too far behind. But poor old X is lagging behind. So, what led to this low ranking for X? Well, it seems that since Elon Musk took over, things have changed, and not for the better. There are unclear policies on climate misinformation and a less communicative content moderation team, both of which have contributed to X’s downward slide in the rankings. Maybe they need to step up their game a bit. In another news story, Google is facing a lawsuit after it allegedly directed a man, Philip Paxson, to drive off a collapsed bridge via Google Maps. Sadly, Paxson lost his life in the tragic accident. According to his family, Google was aware of the bridge’s collapse but failed to update its navigation system, which they argue makes the tech giant negligent in Paxson’s death. Google apparently received reports about the bridge’s condition, but did nothing to fix the route information. It’s a heartbreaking situation. And finally, brace yourself for this one. A study by dappGambl has found that a whopping 95% of NFTs are now practically worthless. Yep, you heard that right. NFTs, which were once all the rage, have lost their shine. Prices have plummeted, and most of the 73,257 NFT collections analyzed have a market cap of zero Ether. It’s uncertain what the future holds for NFTs, but they’ll need to prove their worth, whether through cultural significance or as a representation of actual art, if they want to stick around. So, that’s the latest in tech and climate news. Stay tuned for more updates!
OpenAI has just unveiled their latest model for text-to-image translation called DALL·E 3, and it’s pretty impressive! This new version is built directly on ChatGPT, which means you can use ChatGPT to generate customized and detailed prompts for DALL·E 3. And if the results aren’t exactly what you were hoping for, you can even ask ChatGPT to make some tweaks. Compared to its predecessor, DALL·E 2, DALL·E 3 delivers significant improvements in creating detailed images. OpenAI showcased this by providing a prompt for an expressive oil painting of a basketball player dunking, depicted as an explosion of a nebula. The results from DALL·E 3 were far superior to those from DALL·E 2. OpenAI has also taken steps to ensure that DALL·E 3 doesn’t generate violent, adult, or hateful content. They have designed it to decline requests for images in the style of living artists. Additionally, creators have the option to exclude their images from being used in the training of OpenAI’s future image generation models, giving them more control over the use of their work. Currently, DALL·E 3 is in research preview and will be available to ChatGPT Plus and Enterprise customers in October through the API. It will later be made available in Labs for those interested. This new release is important because it addresses the limitations of previous text-to-image systems, which often ignored certain words or descriptions. With DALL·E 3, AI’s ability to generate images that align precisely with the provided text takes a huge leap forward. It raises questions about how other image generators like Midjourney and Stable Diffusion will keep up. OpenAI has also prioritized safety improvements in DALL·E 3. They have implemented measures to prevent explicit content and have tools in place to identify risky words and block public figures. Furthermore, artists can now request that their work be blocked from AI copying, and DALL·E 3 won’t mimic the styles of specific artists when named. OpenAI hopes that the integration with ChatGPT and the safety guards in DALL·E 3 will expand access to this technology while preventing misuse. However, there are still concerns and legal issues surrounding AI-generated art that need to be addressed.
Amazon had some exciting announcements at its recent devices event. One of the standout updates is the integration of generative AI into their Echo family of devices. This new AI model is optimized for voice, taking into account not only what is said but also body language, eye contact, and gestures. This means that interactions with Alexa will become much more powerful and conversational, providing users with improved experiences. But that’s not all. Amazon has also introduced generative AI updates for Fire TV’s voice search. This update aims to enhance the conversational interaction between users and Alexa, allowing for a more natural and intuitive way to discover new content based on specific preferences. This development is significant because it showcases how integrating language models like Generative AI into voice assistants can revolutionize the way we interact with them. Amazon’s revamp of Alexa using generative AI is a game-changer. It enables voice assistants to better understand context, seamlessly carry over information from previous conversations, and provide a more personalized experience for users. In fact, Amazon is transforming Alexa into a hands-free ChatGPT by leveraging the technology behind chatbots. This upgrade will give Alexa the ability to engage in more complex and open-ended conversations. It will also enhance its simulated personality, interpret body language (for devices with cameras), and modulate its voice for a more natural conversation. However, there are some challenges to overcome, such as responding accurately to body language and refining these large language models to prevent inappropriate or nonsensical responses. But with Amazon’s dedication to improving AI experiences, we can expect significant advancements in these areas. Overall, Amazon’s integration of generative AI into Alexa and Fire TV demonstrates their commitment to providing users with more intuitive, personalized, and conversational experiences.
Hey there! Have you heard about Mark Zuckerberg’s latest philanthropy project? It’s got a pretty ambitious goal – to “cure all diseases”. The project, called the Chan Zuckerberg Initiative (CZI), is a collaborative effort between Zuckerberg and his wife, Priscilla Chan. So here’s the plan: CZI is planning to build one of the biggest GPU clusters in the world specifically for AI-driven biomedical research. They want to use large language models to dive deep into disease development at the cellular level and even predict how cells behave. And to do that, they’re going to need some serious computational power – over 1,000 Nvidia’s H100 GPUs! This high-performance computing system is expected to be up and running by 2024. And let me tell you, it’s going to revolutionize biomedical research. From mapping out various cell types across different organisms to designing potential drugs and therapeutics, this GPU cluster will supercharge the entire process. I don’t know about you, but I’m pretty excited to see what kind of breakthroughs this project will bring. Who knows, maybe we’ll be living in a world where diseases are a thing of the past sooner than we think!
So, let’s dive into the latest AI updates from OpenAI, Microsoft, YouTube, Google, Cisco, and Anthropic. It seems like ChatGPT is back in the spotlight with an increase in usage, particularly because students are returning to school and concerns about AI cheating are on the rise. After experiencing a decline throughout the summer, ChatGPT has seen a 12% traffic increase since fall classes resumed in the US. However, it’s important to note that current usage is still below the peak levels seen earlier this year. With students back in the classroom, concerns about AI-aided cheating have resurfaced. The easier access students have to AI technology raises fresh debates among schools about whether to ban, incorporate, or ignore such tools. For educators, managing responsible AI use in academics is becoming a complex balancing act. There’s also some uncertainty surrounding potential revenue as ChatGPT’s reliance on students could pose challenges for monetization. Moving on to other AI news, Microsoft has announced a new AI-powered feature called Microsoft Copilot. This feature, available in various Windows 11 applications, Microsoft 365, Edge, and Bing, allows users to rearrange windows, generate text, edit pictures, and more. It’s like having Bing integrated into your Windows experience. YouTube is not far behind with its AI advancements. The platform has introduced three new AI-powered features specifically for Shorts creators. Dream Screen uses AI to generate background images and videos, Creator Music helps find the perfect track for Shorts, and AI Insights for Creators assists in brainstorming the next video idea. These features aim to enhance the content creation experience on YouTube. Meanwhile, Google has expanded its AI coding assistant, Studio Bot, to 170 countries. Initially launched for Android developers in the US, this assistant helps generate code, fix errors, and answer questions about Android development. It’s a handy tool for developers worldwide. In the world of image creation, Microsoft’s DALL-E 3 is making its way to Bing. Soon, users will be able to create images in a chat using DALL-E 3. This exciting feature will be rolled out for enterprise users in October, opening up new possibilities for visual communication. Now, let’s switch gears to a significant acquisition. Cisco has announced its plan to acquire cybersecurity firm Splunk for $28 billion. This move aligns with Cisco’s goal to expand its software and AI-powered data analysis capabilities. Splunk, which introduced AI features earlier this year to detect and respond to data anomalies, will play a vital role in Cisco’s strategy. In the realm of responsible AI scaling, Anthropic, the company behind the Claude chatbot, has released a policy that emphasizes its commitment to responsible AI system development. The policy acknowledges the potential for AI systems to cause catastrophic risks, including thousands of deaths or immense financial damage. It’s encouraging to see companies prioritizing responsible AI practices. In other tech news on September 22nd, 2023, Cisco is set to make its largest acquisition ever by acquiring Splunk for $28 billion. This move aims to boost security services and system performance troubleshooting. On a different note, NASA eagerly awaits the return of pristine asteroid Bennu samples, taken by OSIRIS-REx in 2020. The samples could unlock valuable insights into the origins of our solar system. In the legal world, lawyers who sued Tesla’s board for excessive pay are seeking a jaw-dropping $10,000 an hour. The case is sure to attract attention as it unfolds. Another interesting development involves an anonymous developer who used OpenAI’s ChatGPT API to program an AI that created and launched an ERC-20 token called AstroPepeX. Within just 24 hours, the token generated an astonishing $12.9 million in trading. It’s a testament to the possibilities AI offers in the realm of finance and entrepreneurship. Lastly, Ilya Sutskever, one of OpenAI’s renowned figures, along with machine ethicist Thomas Krendl Gilbert, have described AI development as “alchemy.” This comparison underscores the unpredictable and mysterious nature of AI outcomes, sparking heated debate within the industry. And there you have it, the latest AI updates featuring ChatGPT, Microsoft, YouTube, Google, Cisco, and Anthropic. Stay tuned for more exciting advancements in the world of artificial intelligence.
Hey there! If you’re excited about diving deeper into the world of artificial intelligence, I’ve got just the thing for you! There’s this amazing book called “AI Unraveled: Demystifying Frequently Asked Questions on Artificial Intelligence.” Trust me, it’s a game-changer! Now, let me tell you why you should totally get your hands on this gem. “AI Unraveled” is packed with all the answers to those burning questions you may have about AI. Think of it as your ultimate AI guidebook. It’s like having a knowledgeable expert right by your side, unravelling the mysteries of artificial intelligence in a way that’s easy to comprehend. The best part? You can grab a copy of this must-read book at three different platforms: Apple, Google, or Amazon. So, no matter whether you’re an Apple aficionado, a Google guru, or an Amazon enthusiast, there’s a way for you to access this invaluable resource. So, why wait any longer? Dive into “AI Unraveled” today and expand your understanding of artificial intelligence like never before. This book is a game-changer, and it’s ready to be enjoyed by curious minds like yours. Happy reading!
In today’s episode, we covered Microsoft’s AI-powered Copilot, YouTube’s new AI features for creators, evaluating large language models in industry, concerns with AI detection tools in universities, rankings of tech companies tackling misinformation, OpenAI’s DALL·E 3 text-to-image model, generative AI updates from Amazon, Zuckerberg’s philanthropy in AI-driven research, ChatGPT usage concerns, and other notable news – plus, don’t forget to expand your AI knowledge with the essential book ‘AI Unraveled’. Join us next time on AI Unraveled as we continue to demystify frequently asked questions on artificial intelligence and bring you the latest trends in AI, including ChatGPT advancements and the exciting collaboration between Google Brain and DeepMind. Stay informed, stay curious, and don’t forget to subscribe for more!
Microsoft has announced a new AI-powered feature, Microsoft Copilot. It’ll bring AI features into various Windows 11, Microsoft 365, Edge, and Bing. Our first impressions are that it’s Bing but for Windows. You can use Copilot to rearrange windows, generate text, open apps on the web, edit pictures and more.
Copilot can be accessed via an app or with a simple right-click and will be rolled out across Bing, Edge, and Microsoft 365 this fall, with the free Windows 11 update starting on September 26th.
Why does this matter?
While we don’t see any revolutionary use cases of Copilot as of now, it’s still a huge step towards the democratization of AI. As more users get their hands on this AI copilot, we’ll know the true extent of its effectiveness. If all goes well, Microsoft will end up grabbing an even bigger share of the AI market as it will deliver AI natively to all Windows devices.
YouTube announces 3 new AI features for creators
In a YouTube event, the company announced 3 AI-powered features for YouTube Shorts creators.
Dream Screen: It allows users to create image or video backgrounds using AI. All you need to do is type what you want to see in the background and AI will create it for you.
Creator Music: This was a previously available feature but got an AI revamp this time around. Creators can simply type in the kind and length of the music they need and AI will find the most relevant suggestions for their needs.
AI Insights for Creators: This is an inspiration tool which generates video ideas based on AI’s analysis of what the audiences are already watching and prefer.
Why does this matter?
It seems like a strategic decision to natively introduce AI features to support users. It’s a trend we are seeing increasingly more across the landscape. For the users, it’s great news since they get free AI assistance in their creative endeavors.
Google’s innovative approach to train smaller language models
Large language models (LLMs) have enabled new capabilities in few-shot learning, but their massive size makes deployment challenging. To address this, the authors propose a new method called distilling step-by-step, which trains smaller task-specific models using less data while surpassing LLM performance.
First, the key idea is to extract rationales – intermediate reasoning steps – from an LLM using few-shot chain-of-thought prompting. These rationales are then used alongside labels to train smaller models in a multi-task framework, with tasks for label prediction and rationale generation. Experiments across NLI, QA, and math datasets show this approach reduces training data needs by 75-80% compared to standard fine-tuning.
Why does this matter?
This new approach to train smaller models with higher accuracy has the potential to support language models that can be deployed on local devices while retaining the performance that was previously achievable only through LLMs.
4 Crucial Factors for Evaluating Large Language Models in Industry Applications
Based on your end goal, you might fancy one LLM over the other. For instance, some industries value privacy over anything while others might put data accuracy over everything else. In this article, Skanda Vivek shares the 4 critical factors you should always consider when picking a large language model.
He mentions Quality, Economic, Latency, and Privacy to be the 4 resting pillars of your decision. He then goes into details discussing each of these parameters and how you should evaluate a given model against them.
Why does this matter?
The ability to make the right decision when choosing the underlying LLM for your applications is massively important. This article will provide you with valuable insights when it comes to choosing the right LLM.
Some universities are ditching AI detection software amid fears students could be falsely accused of cheating by using ChatGPT
Major universities have discontinued the use of AI detection tools due to concerns about their accuracy, potentially falsely accusing students of cheating with the aid of AI tools like ChatGPT.
AI Detection Tool Concerns
False Accusations of Cheating: Many universities, including Vanderbilt and Northwestern, have stopped using Turnitin’s AI detection tools over worries they might wrongly accuse students of using AI to write essays.
High False Positive Rate: Vanderbilt University highlighted a 1% false positive rate, potentially mislabeling 750 out of 75,000 papers. Similarly, Northwestern University and the University of Texas expressed accuracy concerns, opting not to use the tool.
ChatGPT’s Rise & Challenges
Popularity Among Students: The growing use of ChatGPT by students has educators worried about a surge in academic dishonesty.
Misidentification Issues: A Texas professor mistakenly failed half his class because of false detections by ChatGPT, while other students faced wrongful accusations by anti-plagiarism software.
OpenAI’s Stance
Difficulty in AI Text Detection: OpenAI abandoned its AI text detector due to its low accuracy rate. They’ve also cautioned educators about the unreliability of AI content detectors.
Bias Against Non-English Writers: Many detection tools wrongly labeled content by non-English writers as AI-generated, causing additional concerns.
X ranks lowest in tackling climate misinformation, study reveals
X, formerly known as Twitter, has ranked last in a new assessment by Climate Action Against Disinformation for its management of climate misinformation.
Pinterest scored the highest in terms of addressing climate change misinformation, while other platforms like YouTube, Meta, and Instagram also ranked higher than X.
Changes under Elon Musk’s ownership, including unclear policies on climate misinformation and a less communicative content moderation team, have contributed to X’s low ranking.
Google sued after Maps allegedly directed a man to drive off a collapsed bridge
Philip Paxson, a father of two, died after Google Maps directed him to a collapsed bridge, leading to a fatal car plunge, according to a lawsuit filed by his family.
The family claims Google was informed of the bridge’s collapse but failed to update its navigation system, making the tech company negligent in Paxson’s death.
Despite having received reports about the bridge’s state through its ‘suggest and edit’ feature, Google allegedly took no further actions to correct the route information.
Study finds 95% of NFTs are now worthless
According to a study by dappGambl, 95% of NFTs are now practically worthless, with the majority of the 73,257 NFT collections analyzed having a market cap of zero Ether.
Enthusiasm for NFTs has substantially dropped and prices have plunged, with even hyped-up collections becoming virtually valueless.
The future of NFTs is uncertain; they will need to prove they have inherent value, such as cultural relevance or representing actual art, to survive.
Daily AI Update News from Microsoft, YouTube, Google, Cisco, and Anthropic
A happening day for AI with new AI announcements from Microsoft and YouTube. While Anthropic fears AI’s ‘catastrophic risks.’
Microsoft announces AI Copilot – Microsoft has announced a new AI-powered feature, Microsoft Copilot. It’ll bring AI features into various Windows 11, Microsoft 365, Edge, and Bing. Our first impressions are that it’s Bing but for Windows. You can use Copilot to rearrange windows, generate text, open apps on the web, edit pictures and more.
YouTube brings AI features for creators – YouTube announced 3 new AI-powered features for Shorts creators. Dream Screen uses AI to generate background images and videos. Creator Music uses AI to find the perfect track for Shorts. And AI Insights for Creators helps brainstorm the next video idea.
Google expands AI coding assistant to 170 countries – Google launched Studio Bot in 170 countries. It was previously launched in May for Android developers in the US. The assistant helps devs generate code, fix errors and answer questions about Android.
DALL-E 3 will be available in Bing chat – Microsoft’s recently announced DALL-E 3 will be available in Bing as Microsoft announced users will be able to create images in a chat. DALL-E 3 will be rolled out for enterprise users in October.
Cisco to buy Splunk in $28 billion In its bid to expand software and AI powered data analysis, Cisco announced it will buy cybersecurity firm, Splunk, in $28 billion. Splunk has announced AI features that detect and respond to data anomalies, earlier this year.
Anthropic releases policy on ‘catastrophic risks’ – Anthropic, the company behind Claude chatbot, shared a policy highlighting its commitment to responsible scaling of AI systems. The policy acknowledges AI’s potential to cause “thousands of deaths or hundreds of billions of dollars in damage.”
Other Tech news on September 22nd, 2023
Cisco is set to acquire Splunk for $28 billion to boost security services and system performance troubleshooting, marking its largest acquisition ever.
NASA is eagerly awaiting the return of pristine asteroid Bennu samples, taken in 2020 by OSIRIS-REx, aimed at uncovering the origins of the solar system.
Lawyers who sued Tesla board for excess pay want $10,000 an hour.
An anonymous developer used OpenAI’s ChatGPT API to program an AI to create and launch an ERC-20 token, AstroPepeX, generating $12.9 million in trading within 24 hours.
OpenAI’s Ilya Sutskever and machine ethicist Thomas Krendl Gilbert describe AI development as ‘alchemy’, indicating the unpredictable, mysterious nature of AI outcomes, sparking heated industry debate.
Facebook reveals a slightly darker and subtly tweaked logo under Meta’s refreshed identity system, marking the start of a broader design makeover.
OpenAI has unveiled its new text-to-image model, DALL·E 3, which can translate nuanced requests into extremely detailed and accurate images. Here’s all you need to know:
DALL·E 3 is built natively on ChatGPT, which lets you use ChatGPT to generate tailored, detailed prompts for DALL·E 3. If it’s not quite right, you can ask ChatGPT to make tweaks.
Even with the same prompt, DALL·E 3 delivers significant improvements over DALL·E 2, as shown below (Left: DALL·E 2 results, Right: DALL·E 3). The prompt: “An expressive oil painting of a basketball player dunking, depicted as an explosion of a nebula.”
OpenAI has taken steps to limit DALL·E 3’s ability to generate violent, adult, or hateful content.
DALL·E 3 is designed to decline requests that ask for an image in the style of a living artist. Creators can also opt their images out from training of OpenAI’s future image generation models.
DALL·E 3 is now in research preview and will be available to ChatGPT Plus and Enterprise customers in October via the API and in Labs later this fall.
Why does this matter?
As OpenAI notes, modern text-to-image systems have a tendency to ignore words or descriptions, forcing users to learn prompt engineering. DALL·E 3 represents a leap forward in AI’s ability to generate images that exactly adhere to the text you provide. Will other image generators like Midjourney and Stable Diffusion keep up?
ChatGPT can now generate images
OpenAI revealed the latest iteration of its AI art generator, DALL-E 3 will be coming to ChatGPT Plus and Enterprise members. (Tweet)
DALL-E 3’s New Features
ChatGPT for Prompting*:* Users can have ChatGPT generate descriptive prompts for DALL-E 3 art.
Better Context understanding: The new version follows prompts more precisely than before.:* The new version follows prompts more precisely than before.
Staggered Rollout*:* It will first release to paying ChatGPT users, then for free later but no time frame for free users.
Focus on Safety Improvements
Preventing Explicit Content*:* OpenAI claims robust new safeguards against inappropriate images.
Input Classifiers and Blocklists*:* Tools identify risky words and blocks public figures, so nothing new from the usual CGPT censorship.
Lawsuits Over Copying: DALL-E competitors faced suits alleging use of copyrighted art.
Opt-Out for Artists’ Work*:* Artists can now request their art be blocked from AI copying.
Avoiding Artist Mimicry*:* DALL-E 3 won’t recreate specific artists’ styles when named.
TL;DR: OpenAI hopes new ChatGPT integration and safety guards in DALL-E 3 will expand access and prevent misuse, but legal concerns around AI art persist.
At its annual devices event, Amazon announced a few AI updates:
It will soon use a new generative AI model to power improved experiences across its Echo family of devices. The new model is specifically optimized for voice and will take into account body language as well as a person’s eye contact and gestures for more powerful conversational experiences.
It also introduced generative AI updates for its Fire TV voice search, which promises to bring more conversational ways to interact with Alexa and discover new content based on specifics.
Why does this matter?
Integrating LLMs with voice assistants is a perfect use case. But Amazon’s generative AI revamp for Alexa marks a game-changer. It promises voice assistants that understand context better, carry over information from previous conversations, and become more personalized for users.
Zuckerberg’s philanthropy project is building a massive GPU cluster to ‘cure all diseases’
The Chan Zuckerberg Initiative (CZI), founded by Mark Zuckerberg and his wife Priscilla Chan, plans to build one of the world’s largest GPU clusters for AI-driven biomedical research.
The CZI aims to use large language models to understand disease development at cellular levels and predict cell behaviors, necessitating over 1,000 Nvidia’s H100 GPUs for computational requirements.
The high-performance computing system, expected to be operational in 2024, will accelerate biomedical research, from mapping varied cell types in different organisms to designing potential drugs and therapeutics.
Amazon is turning Alexa into a hands-free ChatGPT
Amazon is upgrading Alexa, its voice assistant, with the technology behind chatbots for more complex and open-ended conversation capabilities.
The new feature, which is still in progress, will show more simulated personality, interpret body language with devices equipped with cameras and modulate its voice for a more natural conversation.
While this advancement holds promise, challenges like responding to body language and the fact that these large language models can sometimes blur out inappropriate or nonsensical things, remains to refine.
ChatGPT Usage is Rising Again as Students Return to School
After declining over the summer, ChatGPT usage has increased, most likely as a result of students returning to class and concerns about AI cheating. (Source)
12% Traffic Increase Last Week: ChatGPT saw a sizable jump in US web traffic as fall classes resumed.
Big Drop Over Summer Break: Traffic declined steadily from May through August when school was out.
Still Below Early 2022 Peaks: But current usage remains below ChatGPT’s peak levels earlier this year. Back to School Brings Old Problems
Cheating Fears Resurface: Easier student access with school back raises fresh concerns about AI-aided cheating.
Schools Still Debating Rules: Many institutions continue deciding whether to ban, incorporate or ignore the technology.
Potential Revenue Uncertainty: Reliance on students could be problematic for monetizing ChatGPT.
With the new school year boosting ChatGPT traffic, managing responsible AI use in academics remains a complex balancing act for educators.
OpenAI launches a red teaming network to make its models more robust
OpenAI has launched the OpenAI Red Teaming Network, a group tasked with aiding risk assessment and mitigation strategies for the company’s AI models.
The Red Teaming initiative aims to catch biases and issues in AI models, enhancing safety and reliability in systems such as DALL-E 2 and GPT-4.
OpenAI calls on experts from diverse fields to participate in the network, emphasizing that confidentiality agreements may apply due to the sensitive nature of the work.
ChatGPT/Bard/Claude one paragraph challenge: “In one paragraph, what is the best use of AI for humanity?”
ChatGPT Plus:
The best use of AI for humanity is to augment human capabilities, address pressing global challenges, and improve the quality of life for all. By analyzing vast amounts of data at unprecedented speeds, AI can aid in medical research, predict and mitigate natural disasters, optimize energy consumption, and personalize education to individual needs. Moreover, it can provide solutions to complex problems, such as climate change, resource scarcity, and public health crises, ensuring a more sustainable and prosperous future for generations to come.
Bard:
The best use of AI for humanity is to use it to solve problems that will make the world a better place for everyone. AI has the potential to improve our lives in many ways, such as developing new and more effective ways to diagnose and treat diseases, creating new renewable energy sources and improving energy efficiency, helping us to adapt to climate change, and reducing poverty and inequality. AI can also help us to better understand ourselves and the world around us. This knowledge can be used to make better decisions about our lives, our communities, and our planet.
Claude:
The best use of AI for humanity is to harness its potential to improve lives and build a better world. With the rapid advances in AI, we have an unprecedented opportunity to tackle major challenges like climate change, disease, inequality, and poverty. If developed responsibly with clear ethical guidelines, AI can be a powerful tool to democratize information, expand access to education and healthcare, drive scientific breakthroughs, increase productivity, and create new industries and jobs. However, we must ensure AI is human-centric, aligning machine capabilities with human values and goals. By complementing human strengths and compensating for weaknesses, AI can enhance human abilities and potential. Whether helping doctors diagnose illnesses, aiding the disabled, optimizing transportation systems, or discovering new materials and energy sources, the possibilities are boundless if we steer AI to augment humanity in a way that benefits all. The technology is not inherently good or bad, but rather its impact depends on how we choose to wield it. With inclusive development and prudent oversight, we can direct AI to create a more just, peaceful, and prosperous world.
Daily AI Update News from OpenAI, Amazon, Google DeepMind, GitHub, Uber, and NVIDIA
Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the day in the world of AI.
OpenAI unveils DALL·E 3 – It understands significantly more nuance and detail than its previous systems. DALL·E 3 is now in research preview and will be available to ChatGPT Plus and Enterprise customers in October via the API and in Labs later this fall. It is built natively on ChatGPT, which lets you use ChatGPT as a brainstorming partner and refiner of your prompts.
Amazon brings generative AI to Alexa and Fire TV – At its annual devices event, Amazon introduced generative AI updates for its Fire TV voice search to bring more conversational ways to interact with Alexa and discover new content. – It will also use a new generative AI model to power improved experiences across its Echo family of devices.
Google DeepMind’s ‘Language Modeling Is Compression’ – This paper views the prediction problem through the lens of compression and evaluates the compression capabilities of large (foundation) models. It shows that LLMs are powerful general-purpose predictors and that the compression viewpoint provides novel insights into scaling laws, tokenization, and in-context learning.
GitHub’s Copilot Chat will now be available to individual users – It is available in public beta for GitHub Copilot individual users in Visual Studio and Visual Studio Code.
Uber Eats to roll out AI-powered assistant – It will help users find deals and explore different food options seamlessly.
NVIDIA to train 50,000 Infosys employees on AI technology – Infosys will set up NVIDIA Centre of Excellence to train and certify employees on NVIDIA’s AI technologies. Also, NVIDIA’s AI Enterprise ecosystem of models, tools, runtimes, and GPU systems will be brought to Infosys’s AI-first offering Topaz.
Amazon recruits 250,000 workers for festive season
Amazon announced its plans to hire 250,000 employees (full-time, part-time, seasonal) in the U.S., for the upcoming holiday rush, a significant increase from last year’s 150,000.
The wages for these new warehouse and delivery roles will range between $17 and $28 per hour, with the average pay now being $20.50, up from $19, and sign-on bonuses worth $1,000 to $3,000 in certain locations.
This announcement is part of a recent trend where Amazon has been raising the starting wages amidst labor tensions, criticism over warehouse injury records, employees’ organization efforts and being the second-largest employer in the U.S., trailing only Walmart.
Google is rolling out Bard’s most capable model yet. Here are the new features:
Bard Extensions in English- With Extensions, Bard can find and show you relevant information from the Google tools you use every day — like Gmail, Docs, Drive, Google Maps, YouTube, and Google Flights and hotels — even when the information you need is across multiple apps and services.
Bard’s “Google it”- You can now double-check its answers more easily. When you click on the “G” icon, Bard will read the response and evaluate whether there is content across the web to substantiate it.
Shared conversations- When someone shares a Bard chat with you through a public link, you can continue the conversation, ask additional questions, or use it as a starting point for new ideas.
Expanded access to existing English language features- Access features such as uploading images with Lens, getting Search images in responses, and modifying Bard’s responses– to 40+ languages.
These features were possible because of new updates made to the PaLM 2 model.
Why does this matter?
These updates make Bard versatile and respond with greater quality and accuracy. The ‘Google it’ button is also a step toward building trust with language models. These strides show Google’s determination to stay competitive in AI, as it is also locked in a race with OpenAI to launch the next generation of multimodal AI models.
Intel’s new chip, due in December, will be able to run a generative AI chatbot on a laptop rather than having to tap into cloud data centers for computing power. It is made possible by new AI data-crunching features built into Intel’s forthcoming “Meteor Lake” laptop chip and from new software tools the company is releasing.
Intel also demonstrated laptops that could generate a song in the style of Taylor Swift and answer questions in a conversational style, all while disconnected from the Internet. Moreover, Microsoft’s Copilot AI assistant will be able to run on Intel-based PCs.
Why does this matter?
This will let businesses test ChatGPT-style AI models without sending sensitive data off their own computers. Intel seems to be on track to become the lead chip manufacturer again, competing with Nvidia to make powerful chips that train AI systems such as ChatGPT and Stability AI’s models.
Google DeepMind’s new system, called AlphaMissense, can tell if the letters in the DNA will produce the correct shape. If not, it is listed as potentially disease-causing.
AlphaMissense can predict the likelihood of genetic diseases by analyzing genetic mutations called missense variants.
AlphaMissense operates like a large language model, trained on human and primate biology, capable of identifying normal sequences of proteins and detecting changes that could suggest a disease.
With 90% accuracy, AlphaMissense is more reliable than existing tools, potentially accelerating the process of identifying disease-causing genetic mutations, which previously required months of meticulous research.
Currently, genetic disease hunters have fairly limited knowledge of which areas of human DNA can lead to disease and have to search across billions of chemical building blocks that make up DNA. They have classified 0.1% of letter changes, or mutations, as either benign or disease-causing. DeepMind’s new model pushed that percentage up to 89%.
Why does this matter?
AI is changing nearly everything we do at the moment and might revolutionize molecular biology and life sciences, too. This development is expected to speed up diagnosis and help search for better genetic disease treatments.
What makes a good AI co-pilot?
Clippy, Siri, and Cortana were here long before the hype for AI copilots began. But their success was limited. Why? (I guess here’s why)
Solving such flawed aspects using LLMs has changed the game around AI assistants.
In this interesting article, Marco Witzmann discusses what AI Assistants (=co-pilots) look like today, what distinguishes them from smart chatbots and not-so-smart speech assistants, as well as a checklist of five properties that separate great from not-so-good co-pilots.
Why does this matter?
Co-pilots enable us to harness the true potential of AI and LLMs. It also helps shape the way we interact with technology and the extent to which AI can assist us in our daily lives.
Meet Notice, the no-code editor designed for user-facing content. Craft wikis, FAQs, blogs, policies, and more using custom AI templates.
Besides creating projects with a single click, you can translate them into over 100 languages and gain insights, all within the same tool.
Notice doesn’t lock your content. As a robust CMS, you can easily publish it to multiple platforms – a website, app, or standalone pages. Plus, enjoy dedicated plugins for WordPress and Shopify.
Elon Musk could charge everyone for X
Elon Musk announced plans for Twitter, now renamed X, to have a small monthly subscription fee to counteract bot activity.
Musk revealed that X currently has around 550 million monthly users, generating between 100 and 200 million posts each day.
Despite criticism over the allowance of hate speech, Musk maintains that X is taking steps to regulate content and manage user verification through a paid system.
Google is turning its Bard AI chatbot into a personal assistant
Google’s Bard AI now has enhanced capabilities, pulling real-time data from Google’s other applications and a user’s data silo to deliver more relevant chatbot responses.
A new feature named Bard Extensions allows the AI to access user’s personal Google data to provide specific answers about their daily activities, while promising not to be used for ad targeting or training the AI model.
To increase transparency and accuracy, Google is introducing a ‘Double Check’ feature where Bard audits its responses and highlights contradictory or heavily referenced statements.
Microsoft AI team accidentally leaks 38TB of private company data
Microsoft’s AI researchers accidentally leaked 38 terabytes of data, including personal passwords, and Teams chat messages due to a misconfigured cloud storage.
Wiz cybersecurity researchers discovered the leak and alerted Microsoft, which soon secured the data and confirmed no unauthorized access occurred.
To prevent future incidents, Microsoft has expanded GitHub’s secret scanning service, a tool that tracks exposed credentials in public open-source code.
This startup is ready to build 10,000 humanoid robots per year
Agility Robotics is preparing to launch its RoboFab, a factory capable of producing over 10,000 advanced humanoid robots, named Digit, annually.
The 70,000-square-foot robot factory, located in Salem, Oregon, is set to open later this year, marking the beginning of mass production of commercial humanoid robots.
Digit is designed to address workforce challenges such as injuries, burnout, high turnover, and labor gaps, with the company’s ultimate goal being to enhance human capabilities.
AlphaMissense, a new model from Google’s artificial intelligence team, analyzes the effects of DNA mutations and will accelerate research into rare diseases.
About 10 years ago, Žiga Avsec was a PhD physics student who found himself taking a crash course in genomics via a university module on machine learning. He was soon working in a lab that studied rare diseases, on a project aiming to pin down the exact genetic mutation that caused an unusual mitochondrial disease.
This was, Avsec says, a “needle in a haystack” problem. There were millions of potential culprits lurking in the genetic code—DNA mutations that could wreak havoc on a person’s biology. Of particular interest were so-called missense variants: single-letter changes to genetic code that result in a different amino acid being made within a protein. Amino acids are the building blocks of proteins, and proteins are the building blocks of everything else in the body, so even small changes can have large and far-reaching effects.
There are 71 million possible missense variants in the human genome, and the average person carries more than 9,000 of them. Most are harmless, but some have been implicated in genetic diseases such as sickle cell anemia and cystic fibrosis, as well as more complex conditions like type 2 diabetes, which may be caused by a combination of small genetic changes. Avsec started asking his colleagues: “How do we know which ones are actually dangerous?” The answer: “Well largely, we don’t.”
Of the 4 million missense variants that have been spotted in humans, only 2 percent have been categorized as either pathogenic or benign, through years of painstaking and expensive research. It can take months to study the effect of a single missense variant.
Today, Google DeepMind, where Avsec is now a staff research scientist, has released a tool that can rapidly accelerate that process. AlphaMissense is a machine learning model that can analyze missense variants and predict the likelihood of them causing a disease with 90 percent accuracy—better than existing tools.
It’s built on AlphaFold, DeepMind’s groundbreaking model that predicted the structures of hundreds of millions proteins from their amino acid composition, but it doesn’t work in the same way. Instead of making predictions about the structure of a protein, AlphaMissense operates more like a large language model such as OpenAI’s ChatGPT.
It has been trained on the language of human (and primate) biology, so it knows what normal sequences of amino acids in proteins should look like. When it’s presented with a sequence gone awry, it can take note, as with an incongruous word in a sentence. “It’s a language model but trained on protein sequences,” says Jun Cheng, who, with Avsec, is co-lead author of a paper published today in Science that announces AlphaMissense to the world. “If we substitute a word from an English sentence, a person who is familiar with English can immediately see whether these substitutions will change the meaning of the sentence or not.”
Pushmeet Kohli, DeepMind’s vice president of research, uses the analogy of a recipe book. If AlphaFold was concerned with exactly how ingredients might bind together, AlphaMissense predicts what might happen if you use the wrong ingredient entirely.
The model has assigned a “pathogenicity score” of between 0 and 1 for each of the 71 million possible missense variants, based on what it knows about the effects of other closely related mutations—the higher the score, the more likely a particular mutation is to cause or be associated with disease. DeepMind researchers worked with Genomics England, a government body that studies the growing pool of genetic data collected by the UK’s National Health Service, to verify the model’s predictions against real-world studies on already-known missense variants. The paper claims 90 percent accuracy for AlphaMissense, with 89 percent of variants classified.
Researchers who are trying to find out whether a particular missense variant may be behind a disease can now look it up in the table and find its predicted pathogenicity score. The hope is that, just as AlphaFold is boosting everything from drug discovery to cancer treatment, AlphaMissense will help researchers in multiple fields accelerate research into genetic variants—allowing them to diagnose diseases and find new treatments faster. “I hope that these predictions will give us an extra insight into which variants cause disease and have other applications in genomics,” says Avsec.
Beyond untangling the effects of single-letter mutations, AlphaMissense demonstrates the potential of AI models in biology more broadly. Because it wasn’t specifically trained to solve the problem of missense variants, but more broadly on what proteins are found in biology, the applications of the model and others like it could reach far beyond single mutations to a better understanding of our whole genome and how it’s expressed—from the recipe book to the whole restaurant. “The basic trunk of the model is derived from AlphaFold,” says Kohli. “A lot of that intuition was, in some sense, inherited from AlphaFold, and we have been able to show that it generalizes to this sort of related but quite different task.”
The researchers stress that the predictions should not be used on their own, but only to guide real-world research: AlphaMissense could help researchers prioritize the slow process of matching genetic mutations to diseases by quickly ruling out unlikely culprits. It could also help improve our understanding of overlooked areas of our genetic code: The model includes an “essentiality” metric for each gene—a measure of how vital it is to human survival. (The function of roughly a fifth of human genes isn’t clear, despite many appearing to be essential.)
Nature – Artificial-intelligence tools that enable companies to share data about drug candidates while keeping sensitive information safe can unleash the potential of machine learning and cutting-edge lab techniques, for the common good.
Researchers have determined how to build reliable machine learning models that can understand complex equations in real-world situations while using far less training data than is normally expected.
OpenAI is working on a Multi-Modal GPT to beat Google Gemini
Both Google and OpenAI are advancing their efforts to develop the next generation of foundation models which will be multimodal.
Google has been making significant progress in this domain with its upcoming Gemini multimodal LLM, which has been introduced to a limited number of external companies.
On the other hand, OpenAI, supported by Microsoft, is diligently working to enhance its GPT-4 model by integrating similar multimodal features. When OpenAI introduced GPT-4 in March, it showcased these features but restricted their availability to a single firm, Be My Eyes, which focuses on technology for visually impaired individuals. Now, OpenAI plans to extend these features, termed GPT-Vision, to a wider audience
OpenAI is also contemplating the development of a subsequent, more advanced multimodal model named Gobi. Unlike its predecessor GPT-4, Gobi’s design emphasizes its multimodal nature from inception.
Llama 2 is an open-source large language model (LLM) developed by Meta and Microsoft. Llama 2 stands for large language model by Meta AI. If you want to understand a large language model, you can visit another blog called What is LLM? Understanding with Examples. Llama 2 is based on the Transformer architecture, which is the same architecture used by other popular LLMs such as GPT-3.
Benefits of Llama 2
Open Source: Llama 2 embodies open source, granting unrestricted access and modification privileges. This renders it an invaluable asset for researchers and developers aiming to leverage extensive language models.
Large Dataset: Llama 2 is trained on a massive dataset of text and code. This gives it a wide range of knowledge and makes it capable of performing a variety of tasks.
Resource Efficiency: Llama 2’s efficiency spans both memory utilization and computational demands. This makes it possible to run it on a variety of hardware platforms, including personal systems and cloud servers.
Scalability: The scalability of Llama 2 signifies its adaptability to larger datasets and its use for more demanding tasks. This makes it a promising tool for the future of Llama 2 research in natural language processing
Easy to use: Llama 2’s accessibility extends to newcomers. Augmented by extensive documentation and a number of tutorials, it fosters ease of use and exploration.
Stability AI launches StableLM, an open-source suite of language models
The creators of Stable Diffusion, Stability AI, just released a suite of open-sourced large language models (LLMs) called StableLM. This comes just 5 days after the public release of their text-to-image generative AI model, SDXL.
StableLM is trained on a new experimental dataset built on The Pile, but three times larger with 1.5 trillion tokens of content. The richness of this dataset gives StableLM surprisingly high performance in conversational and coding tasks, despite its small size of 3 to 7 billion parameters (by comparison, GPT-3 has 175 billion parameters).
Some examples
Stability AI releases a powerful text-to-animation tool
Stability AI released Stable Animation SDK, a tool designed for developers and artists to implement the most advanced Stable Diffusion models to generate stunning animations. It allows using all the models, including Stable Diffusion 2.0 and Stable Diffusion XL. And it offers three ways to create animations:
Text to animation
Initial image input + text input
Input video + text input
The initial image/video inputs act as the starting point for the animation, which is additionally guided by text prompts to arrive at the final output.
Stability AI launches text-to-music AI
Stability AI has launched Stable Audio, a music and sound generation product. Stable Audio utilizes generative AI techniques to provide faster and higher-quality music and sound effects through a user-friendly web interface.
The product offers a free version for generating and downloading tracks up to 45 seconds long and a subscription-based ‘Pro’ version for commercial projects with 90-second downloadable tracks. Stable Audio allows users to input descriptive text prompts and desired audio length to generate customized tracks. The underlying model was trained using music and metadata from AudioSparx, a music library.
Battery prices plummet as electric cars approach ‘tipping point’
The cost of lithium-ion battery cells fell below $100 per kilowatt-hour in August, marking a significant decline in prices that could propel the use of electric vehicles.
This price drop is crucial as energy analysts posit that battery pack prices reaching $100 per kilowatt-hour is the “tipping point” where electric vehicles can compete in price with fossil fuel vehicles.
The decrease in battery cost, contributed by declining lithium prices and discovery of extensive lithium deposits, could also significantly impact other technologies such as solar and wind energy storage.
Intel unveils glass substrates for chips to advance Moore’s Law
Intel has developed a breakthrough glass substrate technology for manufacturing processors, offering higher temperature tolerance and power efficiency, which can result in better performance.
The new technology allows for an increased transistor count, facilitating larger chiplet complexes, faster communication speed, and more bandwidth within the CPU core.
The glass substrate technology, applicable to any processor, might uphold Moore’s Law by boosting processor performance and reducing power requirements, although it’s not expected to be available until later this decade.
Emerging AI Innovations: September 18th, 2023
Mustafa Suleyman, co-founder of DeepMind, believes that we are on the cusp of a new era in artificial intelligence (AI). In what he refers to as the “third wave” of AI evolution, machines will not only communicate with humans but also with other machines. To understand this progression, let’s take a quick look at the previous phases. The initial phase was focused on classification, specifically deep learning algorithms that could classify different types of data. Then came the generative phase, where AI systems used input data to create new information. But now, we’re heading into the interactive phase. This is where machines will be capable of carrying out tasks by conversing not only with humans but also with other AI systems. Users will be able to provide high-level objectives to their AI and let it take the necessary actions, involving dialogue with both machines and individuals. This interactive AI has the potential to be more than just a tool for automation. It will possess the freedom and agency to execute tasks, bringing us closer to the AI we see in science fiction. Instead of being static, it will be dynamic and adaptable, much like the depictions of AI in movies. Interestingly, despite the excitement surrounding generative AI, there seems to be a decline in its popularity. User growth and web traffic for tools like ChatGPT have decreased. DeepMind itself has released a rival to ChatGPT called Pi, which emphasizes its polite and conversational nature. Overall, it’s clear that AI is rapidly advancing, and the future holds great promise for machines that can interact not only with humans but also with their own kind.
So, listen up! Google and DeepMind have been tinkering away to make our Google Maps experience even more personalized. They’ve developed an AI algorithm that suggests routes tailored just for you. I’m talking hyper-personalization here, people. This new algorithm is no joke. It boasts a whopping 360 million parameters and uses real driving data from Maps users to figure out what gets our engines revving when it comes to route decisions. It considers all sorts of factors like travel time, tolls, road conditions, and even our own personal preferences. It’s like having a virtual co-pilot who knows you better than you know yourself. Now, how do they do it? I’m about to drop some serious tech knowledge on you. They use something called Inverse Reinforcement Learning (IRL) to learn from our behavior, and this fancy thing called Receding Horizon Inverse Planning (RHIP) to tackle both short and long-distance travel. Tests have shown that RHIP can suggest routes for two-wheelers with a 16 to 24 percent improvement in accuracy. And here’s the best part: it’s only going to get better over time as it learns more about what routes we prefer. In the past, Google’s attempts to use AI for route planning have hit roadblocks because real-world road networks can be a mind-boggling labyrinth of complexity. But the beauty of RHIP is that it can take on this challenge with a sophisticated approach. It’s proof that better performance is all about scale, both in terms of the data set and the complexity of the model. So, get ready to hit the open road with Google Maps’ hyper-personalized routes, brought to you by the wonders of AI.
So, imagine a world where AI agents play a crucial role in our society. Well, this comprehensive survey on LLM-based agents brings us one step closer to that reality. It’s a deep dive into the world of AI agents and how we can utilize them for the greater good. But what are LLM-based agents, you ask? LLM stands for large language models, and this survey explains why they make a great foundation for AI agents. They present a conceptual framework that can be customized for various applications, making them incredibly versatile. The survey doesn’t stop there; it goes on to explore the numerous applications of LLM-based agents. From single-agent scenarios to multi-agent scenarios and even human-agent cooperation, these agents can play a role in various settings. They even delve into agent societies, examining how LLM-based agents behave and interact with each other. It’s fascinating to see how these agents mirror certain aspects of human society. The survey also highlights key topics and open problems in the field. This is valuable information for developers, as it serves as a practical resource for building AI agents. But it’s not just for developers; researchers, practitioners, and policymakers can also benefit from this survey. It can guide them in further advancing the field of AI and LLM development in a responsible manner. So, why does all of this matter? Well, this survey has the potential to be a game-changer. It offers insights and guidance that could lead to breakthroughs in the world of AI. With responsible development and utilization of LLM-based agents, we can shape a future where humans and AI agents coexist and thrive in harmony.
Hey there! I’ve got some exciting news for all you designers and 3D printing enthusiasts out there. The geniuses over at MIT have come up with an awesome tool called Style2Fab that’s powered by AI and allows you to personalize your 3D-printable models. How cool is that? So here’s the deal: with Style2Fab, you can add custom design elements to your 3D models without messing with the functionality of the objects. All you need to do is describe your desired design using natural language prompts. Yup, you heard it right. No complicated software or coding required. Just good ol’ words to express your creative vision. But wait, it gets even better. Once you’ve described your dream design, you can simply feed it into a 3D printer and bring your creation to life. How awesome is that? This tool really opens up a whole new world of possibilities, especially for those who are just starting out in the design world. But it doesn’t stop there. Style2Fab also has the potential to revolutionize the field of DIY assistive technology and devices. Imagine how clinicians and medical patients could benefit from customized and personalized solutions that are easier to create than ever before. So, folks, get ready to take your 3D printing game to the next level with Style2Fab. It’s time to unleash your creativity and make your designs truly stand out. The future is here, and it’s looking pretty amazing.
Have you ever wondered how many senses AI has? Well, let’s dive into this fascinating topic of multimodal learning to find out! In this article, we’ll explore the next step in AI that’s currently being developed: multimodal learning. Our dear author, Harshvardhan, takes us on a journey to understand how multimodal models work and their potential use cases. Through intriguing analogies, the article sheds light on the technical aspects of multimodal learning and discusses Meta’s efforts in leading open-source research on these models. So, why is this important? By delving into the world of multimodal learning, we gain valuable insights that can spark new applications and research directions. These insights ultimately contribute to the advancement of multimodal AI and its practical applications. Imagine the possibilities we can unlock when AI can truly perceive and comprehend the world through multiple senses! Exciting times lie ahead as we continue to push the boundaries of AI. Multimodal learning opens doors to a future where AI can process and understand information in a more human-like way. Stay tuned for more developments in this groundbreaking field!
In today’s Daily AI News, we have some interesting updates to share. Let’s dive right in! First up, we have news about AI artists being banned by Google. Well, not exactly. Google Colab has actually restricted free users from using the popular Gradio user interface for Stable Diffusion. This decision was made to manage the strain on resources, but users still have options like upgrading to the paid tier or utilizing other free interfaces. Moving on, DeepMind has made a fascinating discovery. They found that large language models (LLMs) can optimize their own prompts using a method called ‘Optimization by PROmpting’ (OPRO). By utilizing ‘meta-prompts,’ LLMs can generate and refine solutions for improved results. This technique can greatly enhance LLM accuracy, but the prompt format is crucial. In other news, MIT researchers have developed a generative AI-driven tool called Style2Fab. This tool allows users to personalize 3D-printable models by adding custom design elements while ensuring the functionality of the objects remains intact. All this can be done through natural language prompts, making it easy and efficient. Next up, Meta is getting ready for the holiday season by launching automated budget scheduling and bid multipliers. These features will help marketers make the most out of their ad campaigns, thanks to AI. SoftBank is also making moves in the AI world. They are considering investing in AI companies, including a potential partnership with OpenAI. The investment could be in the tens of billions, showing the interest in AI’s future. And lastly, Anthropic and BCG have formed an alliance to deliver enterprise AI solutions to clients. This alliance will give BDG’s clients direct access to Claude 2 and Anthropic’s AI technology. According to DeepMind’s cofounder, Mustafa Suleyman, generative AI is just a phase. The future lies in interactive AI. Suleyman envisions building chatbots that can not only chat but also carry out tasks by interacting with other software and people. That wraps up today’s AI news. Stay tuned for more updates and advancements in the exciting world of artificial intelligence!
Hey there! Here are some interesting tech news stories for you. Did you know that the US dominates generative AI funding? They’re drawing in a whopping 89% of global venture capital investment, which amounts to a staggering $20 billion! In contrast, Europe is only getting 5% of that investment, equivalent to $1 billion. OpenAI’s rise seems to be a driving force behind this preference among venture capitalists. In some not-so-great news, it looks like North Korea-linked hackers have allegedly stolen $70 million in crypto assets from CoinEx. Blockchain researchers suspect their involvement in this cyberattack. Moving on to investments, Sequoia and Andreessen’s Instacart investment during the tech boom of 2021 is now facing a bit of a challenge. The company’s upcoming IPO could result in a 75% valuation drop, which is quite significant. Let’s talk about Google now. They’re doing their part to prolong the lifespan of Chromebooks by releasing automatic updates for a whole decade. This move is not only great for saving schools up to $1.8 billion but also helps limit technology waste. Sam Altman, the CEO of OpenAI, seems to be in awe of AI’s success. Despite its global excitement and wide use, Altman acknowledges that there may be challenges ahead, which is an honest and refreshing perspective. That’s all for now! Stay tuned for more tech updates.
Hey there! If you’re excited about diving deeper into the world of artificial intelligence, I’ve got just the thing for you! There’s this amazing book called “AI Unraveled: Demystifying Frequently Asked Questions on Artificial Intelligence.” Trust me, it’s a game-changer! Now, let me tell you why you should totally get your hands on this gem. “AI Unraveled” is packed with all the answers to those burning questions you may have about AI. Think of it as your ultimate AI guidebook. It’s like having a knowledgeable expert right by your side, unravelling the mysteries of artificial intelligence in a way that’s easy to comprehend. The best part? You can grab a copy of this must-read book at three different platforms: Apple, Google, or Amazon. So, no matter whether you’re an Apple aficionado, a Google guru, or an Amazon enthusiast, there’s a way for you to access this invaluable resource. So, why wait any longer? Dive into “AI Unraveled” today and expand your understanding of artificial intelligence like never before. This book is a game-changer, and it’s ready to be enjoyed by curious minds like yours. Happy reading!
In this episode, we explored topics ranging from the future of AI with conversational capabilities, personalized route suggestions in Google Maps, the construction and applications of LLM-based agents, AI tools for personalizing 3D-printable models, advancements in multimodal learning, restrictions on free users and new innovations from Meta, SoftBank’s potential involvement with OpenAI, enterprise AI solutions, interactive AI chatbots, recent news in generative AI funding and cybersecurity, and a recommendation to expand your AI knowledge with the essential book ‘AI Unraveled‘ available at Apple, Google, or Amazon. Join us next time on AI Unraveled as we continue to demystify frequently asked questions on artificial intelligence and bring you the latest trends in AI, including ChatGPT advancements and the exciting collaboration between Google Brain and DeepMind. Stay informed, stay curious, and don’t forget to subscribe for more!
DeepMind co-founder predicts “third wave” of AI: machines talking to machines and people
DeepMind’s co-founder, Mustafa Suleyman, anticipates a “third wave” of AI evolution where machines will interact with both humans and other machines.
The Evolution of AI Phases
Initial Classification Phase: This was the first wave, focusing on deep learning that classifies different types of input data, such as images and audio.
Current Generative Phase: AI uses input data to create new data.
Upcoming Interactive Phase: Machines will be able to perform tasks by conversing with other machines and humans. Users will give high-level objectives to their AI systems which will then take necessary actions, involving dialogues with other AIs and individuals.
Interactive AI’s Potential
More than Just Automation: This AI won’t just be about following commands but will have the freedom and agency to execute tasks.
Closer to Sci-Fi: Interactive AI is anticipated to be more similar to the artificial intelligence depicted in science fiction, with dynamic capabilities rather than being static.
Current AI Landscape:
Generative AI’s Popularity: Despite being a game-changer, enthusiasm for generative AI seems to be waning, with declining user growth and web traffic for tools like ChatGPT.
Inflection AI’s “Pi”: Earlier this year, Suleyman’s company released a ChatGPT rival named Pi, emphasizing its polite and conversational nature.
Google’s AI for hyper-personalized Maps
Google and DeepMind have built an AI algorithm to make route suggestions in Google Maps more personalized. It includes 360 million parameters and uses real driving data from Maps users to analyze what factors they consider when making route decisions. The AI calculations include information such as travel time, tolls, road conditions, and personal preferences.
The approach uses Inverse Reinforcement Learning (IRL), which learns from user behavior, and Receding Horizon Inverse Planning (RHIP), which uses different AI techniques for short- and long-distance travel. Tests show that RHIP improves the accuracy of suggested routes for two-wheelers by 16 to 24 percent and should get better at predicting which route they prefer over time.
Why does this matter?
In the past, Google’s attempts to use AI systems at scale for route planning have often failed due to the sheer complexity of real-world road networks. RHIP can now overcome this hurdle with a sophisticated approach, confirming that better performance is related to scale both in terms of data set and model complexity.
The Rise and Potential of LLM-Based Agents: A survey
Probably the most comprehensive overview of LLM-based agents, this survey-cum-research covers everything from how to construct AI agents to how to harness them for good. It starts by tracing the concept of agents from its philosophical origins to its development in AI and explains why LLMs are suitable foundations for AI agents. It also:
Presents a conceptual framework for LLM-based agents that can be tailored to suit different applications
Explores the extensive applications of LLM-based agents in three aspects: single-agent scenarios, multi-agent scenarios, and human-agent cooperation
Delve into agent societies, exploring the behavior and personality of LLM-based agents, the social phenomena that emerge when they form societies, and the insights they offer for human society
Discuss a range of key topics and open problems within the field
Here’s a scenario of an envisioned society composed of AI agents in which humans can also participate.
Why does this matter?
It is a practical resource for developers to build AI agents. It also serves as a guide for researchers, practitioners, and policymakers to further advancement in the field, potentially leading to breakthroughs in AI and LLM development in a responsible way.
AI makes it easy to personalize 3D-printable models
MIT researchers have developed a generative AI-driven tool that enables the user to add custom design elements to 3D models without compromising the functionality of the fabricated objects. A designer could use this tool, called Style2Fab, to personalize 3D models of objects using only natural language prompts to describe their desired design. The user could then fabricate the objects with a 3D printer.
Why does this matter?
The AI tool empowers novice designers and makes 3D printing more accessible. It could also be used in the emerging area of DIY assistive technology and devices, such as for clinicians and medical patients.
Multimodal Learning
Humans have five senses. How many does AI have?
In this article, Harshvardhan talks about the next step in AI that’s still in the works: multimodal learning. With interesting analogies, the article discusses how multimodal models work, their use cases, and Meta’s efforts in leading open-source research on multimodal models.
Why does this matter?
It gives insights into the technical aspects of multimodal learning, encouraging new applications and research directions and, thus, leading to advancements in multimodal AI and its practical applications. and its practical applications.
Pig kidney functions successfully in human
A pig kidney has functioned inside a human body for about two months, displaying potential for non-human organ transplants.
The success of the xenotransplant, performed at NYU Langone, offers a possible solution to the current organ shortage in the US.
The experimental procedure is still in early stages, with clinical trials subject to FDA approval to ensure safety for conventional human patients.
Apple continues to use our own mortality as marketing
Apple has been incorporating messages about lifesaving features in its marketing, purporting to show that Apple Watch and iPhone can save users from disastrous scenarios.
At the recent iPhone 15 showcase, Apple demonstrated how Watch and iPhone can contribute to life-saving outcomes such as detecting low heart rates, allowing for emergency SOS calls in a blizzard, and detecting a high heart rate in a pregnant woman.
This marketing strategy stands in contrast to previous Apple ads that focused on the enjoyable use of its products and enhances differentiation in a market where functions like fitness tracking and connectivity are no longer unique to Apple.
AI artists banned by Google
Google Colab has not banned Stable Diffusion, but has restricted free users from using the Gradio user interface, a popular remote front-end web UI for Stable Diffusion, due to a surge in its usage causing a strain on resources.
The restriction aligns with Google’s updated terms of service that barred users from running remote UIs and desktops, and users still have alternatives available to them, such as upgrading to the paid tier or utilizing free Stable Diffusion interfaces.
The decision by Google highlights the balance between promoting innovation while managing resources sustainably in the rapidly growing field of artificial intelligence.
Startup could make quantum computers 60x more efficient
French startup Alice & Bob developed a superconducting qubit, inspired by the Schrödinger’s cat experiment, which can potentially make future quantum computers up to 60 times more efficient.
These “cat state” qubits are highly resistant to bit flips but more vulnerable to phase flips, though the latter error is easier to correct.
The company aims to produce a 14-qubit system by the end of 2023 and will require an additional six months for calibration.
TikTok is tracking employee back-to-office attendance with new internal tool
TikTok has introduced a new internal tool named MyRTO to track employee office attendance.
The MyRTO system monitors badge swipes when employees enter the office, requiring explanation for any “deviations” from expected attendance.
The move comes amid a broader trend of employee monitoring, especially after the widespread shift to remote work due to the pandemic.
Daily AI Update News from Google, MIT, Meta, SoftBank, Anthropic, BCG, and DeepMind
Google’s AI for hyper-personalized Maps routes – Google and DeepMind have built an AI algorithm to make route suggestions in Google Maps more personalized. It includes 360 million parameters and uses real driving data from Maps users to analyze what factors they consider when making route decisions. – The approach uses Inverse Reinforcement Learning (IRL), which learns from user behavior, and Receding Horizon Inverse Planning (RHIP), which uses different AI techniques for short- and long-distance travel. Tests show that RHIP improves the accuracy of suggested routes for two-wheelers by 16 to 24 percent.
The rise and potential of LLM-based Agents – Probably the most comprehensive overview of LLM-based agents, this survey covers everything from how to construct these agents to how to harness them for good. It also delves into the origin of AI agents, their fit with LLMs, and outlines a framework that can be tailored to suit different applications.
AI makes it easy to personalize 3D-printable models – MIT researchers developed a generative AI-driven tool that enables the user to add custom design elements to 3D models without compromising the functionality of the fabricated objects. A designer could use this tool, called Style2Fab, to personalize 3D models of objects using only natural language prompts to describe their desired design. The user could then fabricate the objects with a 3D printer.
Meta is prepping world’s first AI-powered holiday season – It is prepping marketers for an AI holiday season with the launch of automated budget scheduling and bid multipliers to help make the most out of ad campaigns.
SoftBank considers investment or partnership with OpenAI – It is seeking to invest in AI companies in tens of billions. It is considering various options, including a strategic partnership with, or investment in, Open AI.
Anthropic and BCG form a new alliance to deliver enterprise AI to clients – This will give BDG’s clients direct access to Claude 2 and Anthropic’s AI tech for use across different strategic solutions.
Generative AI is just a phase. What’s next is interactive AI, says DeepMind’s cofounder – Mustafa Suleyman wants to build a chatbot that does much more than chat. Bots that can carry out tasks you set for them by calling on other software and other people to get stuff done.
Emerging AI Innovations: September 17th, 2023
DeepMind discovers that AI large language models can optimize their own prompts
DeepMind introduced a new method called ‘Optimization by PROmpting’ (OPRO), allowing large language models (LLM) to optimize their own prompts using natural language.
The OPRO process utilizes ‘meta-prompts’ to guide the LLM in generating solutions, evaluating and refining solutions for improved results.
This technique can significantly boost LLM accuracy, but outcomes heavily rely on the prompt format, highlighting the intricacy of language models.
EV sets new record for longest distance driven on a single charge
Students at the Technical University of Munich built an electric car “muc022”, which traveled almost 1,600 miles on a single charge, setting a new Guinness World Record.
The car took 99 hours to cover this distance, reaching a top speed of 26 mph, the car was designed to minimize weight and air resistance, and weighs only 375 pounds.
This feat shattered the previous record of 999 miles set by IT Asset Partners in 2017 and far exceeds the longest-range production car, the 2023 Lucid Air, which has a range of 516 miles.
US dominates in generative AI funding, drawing 89% ($20B) of global VC investment, leaving Europe with only 5% ($1B), driven by OpenAI’s rise in VCs’ preference.
North Korea-linked hackers are suspected of stealing $70 million in crypto assets from CoinEx, according to blockchain researchers.
Sequoia and Andreessen’s Instacart investment, made during 2021’s tech boom, faces a 75% valuation drop with Instacart’s upcoming IPO.
Google is prolonging Chromebook lifespan by releasing automatic updates for a decade, a measure that could save schools $1.8 billion and limit technology waste.
OpenAI CEO, Sam Altman, expresses shock at AI’s success and admits to possible challenges ahead, despite its wide use and global excitement.
Reddit launched an AI-powered keyword research tool that will help advertisers.
Infosys is likely to collab with NVIDIA to train 3 lakh+ employees on AI!
India’s Reliance partners with Nvidia to develop a new LLM.
Researchers at Humboldt University in Berlin have developed a biased GPT model called OpinionGPT.
Nasdaq has received SEC approval for its first exchange AI-powered order type!
Instagram might be getting generative AI panoramas.
IRS will focus on the wealthy, using AI to identify sophisticated schemes to avoid taxes.
YouTube announces AI-powered creative guidance in Google Ads
AI chatbots tasked to run a tech company built software in 7 minutes for less than $1.
A boy saw 17 doctors over 3 years for chronic pain. ChatGPT found the right diagnosis.
Roblox, a popular online gaming platform, is set to introduce an AI chatbot called Roblox Assistant.
Alibaba made its AI model ‘Tongyi Qianwen’, available to the public.
Coca-Cola has used AI to create a new flavor called Y3000 Zero Sugar.
15 US companies have taken an oath to develop AI responsibly.
Character.ai outperforms ChatGPT in mobile app usage within the US.
Pika Lab have Introduced Camera Movement Parameter.
Amazon has launched gen AI capabilities to help sellers write more engaging and effective product descriptions.
Hugging Face has launched Würstchen, a diffusion model for image generation.
Researchers have developed an AI tool called RETFound that can diagnose and predict the risk of various health conditions.
Parle Products, the biscuit brand in India, has partnered with IBM to drive its digital transformation using cloud and AI.
EY has invested $1.4 billion in AI and is launching a platform that offers AI-powered versions of its consulting products.
OpenAI is opening its first European Union (EU) office in Dublin.
AWS partnering with India’s ISRO to boost AI capabilities in the space via cloud computing.
Microsoft has open-sourced EvoDiff, a protein-generating AI framework.
Data analytics and AI software maker Databricks has raised over $500 million in a Series I funding round, increasing its valuation to $43 billion.
Infosys, India’s second-largest software services exporter, has signed a $1.5 billion contract to leverage AI solutions.
Emerging AI Innovations: September 16th, 2023
Google’s new DeepMind Maps algorithm improves route suggestions by 24%
Google has developed an AI algorithm to refine route suggestions on Google Maps, personalizing it based on user data and behavior, allegedly improving the accuracy on an average by 16-24 percent.
Personalized Route Suggestions through AI
The AI model comprises 360 million parameters, using real-time data from Maps users to influence factors including travel time, road conditions, tolls, and personal preferences to suggest routes.
This technology is grounded on “inverse reinforcement learning” (IRL), specifically a new IRL algorithm – “Receding Horizon Inverse Planning (RHIP)”.
The Power of RHIP and AI in Maps
Google and Deepmind jointly worked to develop RHIP, using complex stochastic models in immediate vicinity areas, but switching to simpler deterministic methods for distant areas for power conservation.
The AI improves route suggestions for both driving and two-wheeled vehicles by learning from Maps users’ movements and behaviors over time.
Google states that this is the largest application of inverse reinforcement learning for route planning to date.
Implementation and User Testing
Google has applied the algorithm to Maps data globally, but extensive user testing is needed to confirm if the technique consistently produces better routes.
Previous attempts at using AI systems for route planning on a large scale have often failed due to the complexity of road networks.
Misinformation spread: AI systems can propagate misinformation, but research indicates humans are more responsible for spreading falsehoods.
Existential threats: Concerns about AI causing potential extinction events are highlighted, but humans are more likely to initiate catastrophic events like nuclear wars.
Human errors: AI might be portrayed as a threat in scenarios like cybersecurity, but human errors are often the culprits behind major security breaches.
Domains Where AI Outperforms Humans:
Traffic safety: Human error causes 98% of accidents, with autonomous cars predicted to reduce this significantly.
Medical diagnosis: Machine performance is increasingly surpassing human doctors in specific areas.
Creativity & Legal contexts: AI has shown superior performance in art authentication, correcting wrongful human convictions, and reducing trial times.
Algorithmic Bias & Cybersecurity:
Bias in algorithms: Algorithmic bias can be a concern, but it is more easily fixed than human bias. Algorithms, when trained on biased human data, will reproduce this bias.
Cyber threats: Most high-profile cyberattacks are initiated by humans, and AI serves as a defense against such threats, detecting human coding mistakes and correcting them.
NVIDIA’s new software boosts LLM performance by 8x
NVIDIA has developed a software called TensorRT-LLM to supercharge LLM inference on H100 GPUs. It includes optimized kernels, pre- and post-processing steps, and multi-GPU/multi-node communication primitives for high performance. It allows developers to experiment with new LLMs without deep knowledge of C++ or NVIDIA CUDA. The software also offers an open-source modular Python API for easy customization and extensibility.
(The following figures reflect performance comparisons between an NVIDIA A100 and NVIDIA H100.)
Additionally, it allows users to quantize models to FP8 format for better memory utilization. TensorRT-LLM aims to boost LLM deployment performance and is available in early access, soon to be integrated into the NVIDIA NeMo framework. Users can apply for access through the NVIDIA Developer Program, with a focus on enterprise-grade AI applications.
Amazon, Nvidia, Microsoft, and Google lead hiring surge in GenAI
There is an explosive demand for Generative AI talent today. Here are some compelling statistics.
The number of companies mentioning “Generative AI” in monthly job postings is increasing exponentially.
Tech giants leading the surge in hiring for GenAI talent include Amazon, Nvidia, Oracle, Microsoft, Google, and more. Big banks like Citigroup and CapitalOne are also hiring big in GenAI.
Unsurprisingly, technology is the #1 sector looking to hire GenAI experts. Finance is #2nd, and healthcare is #3, while demand has been tepid in sectors like real estate, basic materials, and energy.
Companies are paying a lot for GenAI talent! Among all technical skills/technologies tracked, jobs mentioning “Generative AI” or “LLMs” had the highest average base salary offered, with an average of $200,837/year.
Meta confirms: No ads coming to WhatsApp
Meta has contradicted a Financial Times report suggesting that it plans to display ads on WhatsApp, stating there are no tests or plans for such an initiative.
The company currently monetizes WhatsApp through WhatsApp Business, a service for merchants boasting over 200 million monthly active users, rather than through placing ads.
Despite speculation and past considerations around integrating ads on WhatsApp, these plans have always been cancelled, with a focus instead on refining business and payment services on the platform.
Some helpful AI tools or applications that can enhance your work
1. ChatGPT, Bing Chat, and Google Bard:
These popular chatbots assist with text-related tasks, answering queries, generating content ideas, and more. They can become virtual assistants for tasks like social media posts, code debugging, and proofreading.
2. Tuberank Jeet:
For YouTubers, this tool optimizes videos for YouTube’s algorithm, aiding in SEO-friendly titles, descriptions, and tags.
3. Sendster:
This email marketing tool uses AI to create newsletters, subject lines, headlines, and complete email copies quickly, streamlining email campaigns.
4. VidScribe AI:
It automates subtitle and voiceover creation for videos in multiple languages, saving time on translation.
5. Bing Image Creator and MidJouney:
AI-generated visuals are handy for social media and blogs, reducing the need for graphic design skills.
6. AI Collective:
A versatile AI script that combines different AI language models into one platform, saving time and money by consolidating content creation tools.
AI, particularly generative AI, is transforming work across sectors. These tools are just a glimpse of how AI can enhance productivity and creativity in our daily tasks.
Daily AI News 9/16/2023
A little boy named Alex saw 17 different doctors over the course of three years, unable to find a root cause of his chronic pain. At her wit’s end, his mom, Courtney, fed his radiology report into ChatGPT and produced immediate answers.[1]
In January, Wharton professor Christian Terwiesch gave his MBA final exam to ChatGPT. It passed with flying colors. Now, he’s at it again with a new experiment to determine whether ChatGPT can come up with product ideas better and faster than his students. It can. And cheaper, too.
Bathroom-cleaning robot built for commercial businesses gives consumers hope for AI maid.
Judge admits he used ChatGPT to write a Court of Appeal ruling as he calls the AI tool ‘jolly useful’.
Controversy as ancient human fossils sent to space
Richard Branson’s Virgin Galactic has sparked controversy by sending ancient hominin bones into space, drawing criticism from the paleontology community who view it as a disrespectful stunt.
This marks the first time human ancestral remains have been sent into space, with the included bones belonging to the two-million-year-old Australopithecus sediba and 250,000-year-old Homo naledi, both discovered in South Africa.
Despite being legal, the act of white scientists selecting and sending African fossils into space has been viewed by some as a perpetuation of past unethical practices in palaeoanthropological research.
Elon Musk’s X finally agrees to try and settle Twitter’s mass layoffs lawsuit
Elon Musk’s company, X, recently laid off two-thirds of its workforce, alleging huge daily losses, which led to multiple lawsuits including lack of severance pay and targeting of women.
X has now agreed to attempt to settle these lawsuits, represented by Shannon Liss-Riordan, an attorney who is advocating for nearly 2,000 former employees.
The further schedule for the settlement remains unclear, however negotiations could take place in early December.
Water safety is a critical concern for municipal water systems, as contamination by bacteria and harmful particles can have severe health repercussions. Unfortunately, detecting these issues can be challenging before they cause health problems. To address this need, Clean Water AI has developed an innovative solution that leverages artificial intelligence (AI) to identify water contamination. By utilizing trained models, Clean Water AI’s system can effectively recognize harmful particles and bacteria that may compromise water safety. The solution involves the implementation of distributed devices that continuously monitor water sources for any signs of contamination. These devices are equipped with AI algorithms, which allow them to detect and classify dangerous bacteria and particles accurately. This real-time monitoring enables cities to identify and respond to contamination issues promptly. Clean Water AI employs a deep learning neural network to detect bacteria and particles in water, even at the microscopic level.
By training a convolutional neural network model on the cloud, the system gains the capability to accurately identify and classify various contaminants. To deploy the solution, Clean Water AI utilizes edge devices equipped with the trained model. This approach ensures that the classification and detection occur at the source, providing real-time analysis of water quality. The system is designed to run continuously, allowing for round-the-clock monitoring. Implementing the solution involves the installation of Internet of Things (IoT) devices across different water sources in cities. These devices serve as the frontline sensors, constantly monitoring water quality and detecting any signs of contamination. This comprehensive monitoring approach offers cities greater visibility into their water systems and enables them to take proactive measures to ensure public safety. Clean Water AI has already built a proof of concept using a microscope and Up2 board, keeping the costs under $500. With plans to scale up production, the team aims to reduce unit costs further, making the technology more accessible and affordable for widespread adoption. By leveraging AI and IoT technologies, Clean Water AI offers an effective and efficient solution to address the challenges associated with maintaining water safety in municipal systems. Their innovative approach provides continuous, real-time monitoring, allowing for swift intervention and better safeguarding of public health.
Microsoft Research has developed a novel method known as RAIN to address the challenge of aligning language models with human preferences. This method eliminates the need for fine-tuning or additional data by incorporating self-evaluation and rewind mechanisms. Through a process of self-boosting, unaligned models can generate responses that align with human preferences. One notable feature of RAIN is that it operates without training or parameter updates. Instead, it employs a fixed-template prompt to guide the model’s alignment with users’ preferences. Experimental results have shown that RAIN significantly improves the harmlessness rate of language models while preserving their helpfulness. This methodology also establishes a new defense baseline against adversarial attacks. The implications of RAIN are significant for user safety. By enabling language models to align with human preferences, it reduces the occurrence of harmful outputs and promotes more helpful responses across various applications. Whether it is in the context of customer support or content generation, RAIN enhances the overall user experience by ensuring that language models prioritize safety and comply with user preferences.
Google Research has introduced a new method for transforming single still images into seamless looping videos or interactive dynamic scenes. This method involves training a model on real video sequences that exhibit natural motion, like trees swaying or clothes blowing in the wind. Using this trained model, it becomes possible to predict long-term motion patterns in the Fourier domain based on a single image. These predictions can then be transformed into dense motion trajectories. These motion trajectories have various applications, such as generating dynamic videos from still images or enabling realistic interactions with objects in images. This research is significant because it enhances user experiences by enabling the creation of dynamic videos from static images and facilitating realistic interactions. Furthermore, it has broader implications in the fields of computer vision and artificial intelligence. This technology can potentially find applications in areas like robotics and autonomous systems. By leveraging the power of generative image dynamics, Google Research is pushing the boundaries of what is possible with still images. This breakthrough opens up new possibilities for creative expression and practical applications, ultimately enriching our visual experiences in the digital realm.
Google is set to challenge OpenAI’s GPT-4 model with its own conversational AI software called Gemini. Aimed at providing competition to OpenAI, Gemini consists of a range of large-language models that can perform various tasks such as powering chatbots, summarizing text, generating original content, assisting with code writing, and creating images based on user requests. Although Google is currently granting developers access to a version of Gemini, the largest version of the software is still under development. The company plans to make Gemini available to businesses through its Google Cloud Vertex AI service. This move indicates Google’s strong investment in generative AI as it strives to catch up with OpenAI’s ChatGPT. The significance of this development lies in the potential for more efficient customer support, faster content creation, and improved code development. Google’s Gemini holds the promise of advanced conversational AI, resulting in more powerful chatbots, ultimately enhancing the user experience across various applications. In summary, Google’s forthcoming Gemini software represents the company’s endeavor to rival OpenAI’s GPT-4 model by providing developers with access to a collection of large-language models. This breakthrough has the potential to revolutionize customer support, content creation, and code development, ultimately improving the user experience in multiple domains.
AI models have reached a remarkable level of accuracy in predicting how US judges will rule, even without considering the facts of the case. By analyzing various datapoints such as the judge’s educational background, net worth, rulings based on the origin of lawyers, and their experience in public law, private practice, and state judgeships, these models have achieved an impressive 86% accuracy rate. This raises important questions about the US justice system itself. One cannot help but wonder what implications this has for the fairness and objectivity of judicial decisions. If demographic information alone can predict a judge’s ruling with such high accuracy, it casts doubt on whether the facts of a case truly drive the outcome. It suggests that certain biases and predispositions may subtly influence judicial decisions, even when consciously disregarding the case details. In a different sphere, we find that employees are experiencing what is being referred to as ‘AI Anxiety.’ This is revealed in a LinkedIn report, which shows that almost half of the respondents feel they lack sufficient knowledge about AI. Moreover, nearly 40% admitted to pretending to be more knowledgeable about AI than they actually are, in order to appear informed among their colleagues. Additionally, there is a concern about potential job disruptions due to the rise of AI. A Goldman Sachs study suggests that generative AI could potentially disrupt over 300 million jobs globally. Administrative workers and women seem to be at the highest risk of being replaced by AI. However, this shift also brings about a growing demand for AI skills, with job postings mentioning GPT or ChatGPT experiencing a significant increase. While the usage of AI tools like ChatGPT is prevalent, it is surprising to learn that a majority of users (70%) do not inform their superiors about its use, as highlighted in a Fishbowl survey. Nonetheless, despite the concerns surrounding AI, the traffic to ChatGPT’s website has been declining for three consecutive months. All these developments indicate a complex and evolving relationship between AI and human society. As AI continues to advance, it is crucial for individuals and organizations to address the knowledge gaps and prepare for potential changes in the job market. It is equally important for society as a whole to critically examine the influence of AI on fields like law and ensure that fairness and transparency are upheld.
Google is preparing to release its conversational AI software, Gemini, which is designed to compete with OpenAI’s GPT-4 model. Gemini is a collection of language models (LLMs) that can be used to power chatbots, summarize text, generate original text, assist with code writing, and create images based on user requests. Although developers currently have access to a version of Gemini, the largest version is still being developed. Google plans to make Gemini available to companies through its Google Cloud Vertex AI service. OpenAI is establishing its first office in the European Union (EU), specifically in Dublin, and is making strategic hires in preparation for regulatory challenges. The company is currently recruiting for various positions focused on legal, privacy, and media relations. OpenAI aims to demonstrate its commitment to privacy and comply with EU regulations, particularly in light of previous scrutiny over its ChatGPT chatbot. This move by OpenAI reflects the AI industry’s recognition of the EU’s importance. A new paper from Google Research introduces a model that prioritizes scene dynamics in image spaces. This model is trained on real video sequences that exhibit natural motion, such as swaying trees or blowing clothes. Given a single image, the model can predict long-term motion patterns in the Fourier domain. These predictions can be converted into dense motion trajectories, enabling various applications like generating dynamic videos from still images or facilitating realistic interactions with objects in pictures. Amazon Web Services (AWS) has partnered with India’s ISRO and IN-SPACe to enhance AI capabilities in the space sector through cloud computing. This collaboration aims to support space-tech innovations and empower startups, research institutes, and students. Cloud computing will enable quicker decision-making and push the boundaries of what’s possible in the space industry. The partnership will provide access to state-of-the-art cloud technologies, streamline the management of space data, and support AI, machine learning (ML), and analytics workloads. AWS will also support eligible space startups through the AWS Activate program and offer expertise in building aerospace and satellite solutions via the AWS Space Accelerator program. Microsoft has open-sourced EvoDiff, an AI framework for generating proteins. The process of designing proteins in the lab is currently expensive and time-consuming. EvoDiff aims to simplify this process by generating high-quality proteins without requiring structural information about the target protein. The framework has been trained on data from various species and functional classes of proteins. Databricks, a data analytics and AI software maker, has raised over $500 million in a Series I funding round, which has increased its valuation to $43 billion. Investors in this round include T. Rowe Price, Morgan Stanley, Fidelity, Franklin Templeton, Capital One Ventures, Nvidia, Andreessen Horowitz, and Tiger Global. Infosys has signed a $1.5 billion contract to utilize AI solutions over a 15-year period. While the client’s name has not been disclosed, Infosys will provide enhanced digital experiences and business operation services using its platforms and AI solutions. SpaceX’s Starlink satellite internet service reported $1.4 billion in revenue for 2022, significantly lower than Elon Musk’s earlier projection of $12 billion. The company achieved its first profit in 2023 but fell short of its customer acquisition goals, ending the year with only one million active subscribers instead of the forecasted 20 million. Additionally, Starlink’s reliability was called into question by the FCC, resulting in the reversal of a decision to grant the company $885.5 million in funds. Unity Engine has eliminated its royalty-free licensing structure and will now charge developers on a “per-install” basis once certain thresholds are met. The new Unity Runtime Fee, effective from January 1, 2024, will impose different costs per install depending on the subscription tier. This change has sparked outrage among game developers who embraced Unity for its previous royalty-free structure. Amazon has introduced an artificial intelligence tool for sellers to assist in creating engaging product page content. The AI tool can generate product titles, descriptions, and listings, potentially enhancing seller success. This effort to utilize AI aligns with Amazon’s broader strategy of incorporating AI applications throughout its business. EY has launched EY.ai, a comprehensive platform aimed at facilitating clients’ adoption of AI. This platform has been developed in collaboration with major companies like Microsoft and IBM, with EY investing $1.4 billion into its development. EY.ai integrates AI into existing EY technologies and includes a large language model called EY.ai EYQ. EY’s Chief Technology Officer, Nicola Morini Bianzino, emphasized that these proprietary AI offerings provide clients with confidence and a roadmap for successful AI adoption.
Looking to deepen your knowledge of artificial intelligence? Look no further than “AI Unraveled: Demystifying Frequently Asked Questions on Artificial Intelligence.” This essential book is now available for purchase through various platforms, including Amazon, Barnes and Nobles, Apple, Google, and Amazon today! With “AI Unraveled,” you can expect a comprehensive exploration of artificial intelligence, designed to demystify this complex field. Written by experts in the industry, this book addresses frequently asked questions, making it accessible to both beginners and more seasoned AI enthusiasts. Whether you’re curious about the potential applications of AI, its impact on various industries, or its ethical considerations, this book covers it all. It provides a thorough understanding of AI’s inner workings, enabling readers to grasp concepts such as machine learning, neural networks, and natural language processing. By delving into real-world examples and case studies, “AI Unraveled” bridges the gap between theory and practice. It equips readers with the knowledge they need to navigate the AI landscape confidently. So, if you’re ready to unravel the mysteries of artificial intelligence, secure your copy of “AI Unraveled: Demystifying Frequently Asked Questions on Artificial Intelligence” today. Expand your understanding and stay ahead in this rapidly evolving field. Get your copy now!
In this episode, we explored a range of exciting AI developments, including Clean Water AI’s affordable contamination detection using IoT devices, Microsoft Research’s RAIN method for aligning language models with human preferences, Google’s dynamic video creation from still images, Google’s development of Gemini conversational AI to rival OpenAI’s GPT-4, the accuracy of AI models in predicting US judges’ rulings, and various other updates such as OpenAI’s Dublin office, Microsoft’s open-sourced EvoDiff, Databricks’ funding, Infosys’ AI contract, and more, along with the availability of the book “AI Unraveled: Demystifying Frequently Asked Questions on Artificial Intelligence.” Join us next time on AI Unraveled as we continue to demystify frequently asked questions on artificial intelligence and bring you the latest trends in AI, including ChatGPT advancements and the exciting collaboration between Google Brain and DeepMind. Stay informed, stay curious, and don’t forget to subscribe for more!
Clean Water AI – Summary
Clean Water AI uses a deep learning neural network to detect dangerous bacteria and harmful particles in water. Drinking water can be seen at a microscopic level with real-time detection.
Clean Water AI trains the convolutional neural network model on the cloud, then deploys it to edge devices. We used Caffe, a deep learning framework, which allows a higher frame rate when running with Intel Movidius Neural Computing Stick.
An IoT device can then classify and detect dangerous bacteria and harmful particles. The system can run continuously in real time. The cities can install IoT devices across different water sources to monitor water quality as well as contamination in real time.
Currently, Clean Water AI has been built as a proof of concept using a microscope and Up2 board. The entire prototype costs less than $500, and they’re plans to scale up production to help reduce unit costs.
The paper introduces a method called RAIN that allows language models to align themselves with human preferences without the need for finetuning or extra data. By integrating self-evaluation and rewind mechanisms, unaligned models can produce responses consistent with human preferences through self-boosting.
RAIN operates without training or parameter updates and uses a fixed-template prompt to guide the model’s alignment with human preferences. Experimental results show that RAIN significantly improves the harmlessness rate of language models while maintaining their helpfulness. It also establishes a new defense baseline against adversarial attacks.
Why does this matter?
RAIN enhances user safety by allowing language models to align with human preferences, reducing harmful outputs and ensuring more helpful responses in various applications, from customer support to content generation.
Google Research’s new generative image dynamics
Google Research’s new paper introduces a method for turning single still images into seamless looping videos or interactive dynamic scenes. The model is trained on real video sequences with natural motion, such as trees swaying or clothes blowing in the wind.
Given a single image, the model can predict long-term motion patterns in the Fourier domain. These predictions can be converted into dense motion trajectories, which can be used for various applications, such as creating dynamic videos from still images or enabling realistic interactions with objects in pictures.
Why does this matter?
This research enhances user experiences by enabling dynamic videos from still images and realistic interactions. It can also can have broader applications in computer vision and AI, including robotics and autonomous systems.
Google Challenges GPT-4 with Gemini
Google is reportedly nearing the release of its conversational AI software, Gemini. Which is intended to compete with OpenAI’s GPT-4 model. Gemini is a collection of large-language models that can power chatbots, summarize text, generate original text, help write code and create images based on user requests.
Google is currently giving developers access to a version of Gemini, but not the largest version it is developing. The company plans to make Gemini available to companies through its Google Cloud Vertex AI service. Google has invested heavily in generative AI to catch up with OpenAI’s ChatGPT.
Why does this matter?
Imagine more efficient customer support through smarter chatbots, faster content creation, and enhanced code development. As Google’s Gemini promises advanced conversational AI, benefiting users with more powerful chatbots, which will enhance the user experience in various applications.
AI models can now predict how a US judge will rule with 86% accuracy — without even considering the facts of the case.
Datapoints include where the judge went to law school, what their net worth is, how they rule when the lawyers are from big law firms versus boutique practices and the judges’ history in public law, private practice and state judgeships.
If you can predict how the judge will rule 86% of the time without the facts of the case, primarily based on demographic information, what does that say about the US justice system?
Two-minutes Daily AI Update News from Google, OpenAI, Google Research, AWS, Microsoft, Databricks, and Infosys
Google is reportedly nearing the release of its conversational AI software, Gemini. – Gemini is intended to compete with OpenAI’s GPT-4 model. Gemini is a collection of LLMs that can power chatbots, summarize text, generate original text, help write code, and create images based on user requests. – Google is currently giving developers access to a version of Gemini, but not the largest version it is developing. – The company plans to make Gemini available to companies through its Google Cloud Vertex AI service.
OpenAI is opening its first office in the European Union (EU) in Dublin and making strategic hires as it prepares for regulatory challenges. – The company is currently hiring for 09 positions in Dublin, including roles focused on legal, privacy, and media relations. – OpenAI aims to demonstrate its commitment to privacy and comply with EU regulations, particularly in light of previous scrutiny over its ChatGPT chatbot. – OpenAI’s move to establish a presence in the EU reflects the region’s importance in the AI industry.
Google Research’s new paper introduces a modeling an image-space prior on scene dynamics. – The model is trained on real video sequences with natural motion, such as trees swaying or clothes blowing in the wind. Given a single image, the model can predict long-term motion patterns in the Fourier domain. – These predictions can be converted into dense motion trajectories, which can then be used for various applications, such as creating dynamic videos from still images or enabling realistic interactions with objects in pictures.
Amazon’s AWS has partnered with India’s ISRO and IN-SPACe to enhance AI capabilities in the space sector through cloud computing. – The collaboration aims to support space-tech innovations and empower startups, research institutes, and students. – Cloud computing will enable quicker decision-making and push the boundaries of what’s possible in the space industry. – The partnership will provide access to state-of-the-art cloud technologies, streamline the management of space data, and support AI, ML, and analytics workloads. – AWS will also support eligible space startups through the AWS Activate program and offer expertise in building aerospace and satellite solutions through the AWS Space Accelerator program.
Microsoft has open-sourced EvoDiff, a protein-generating AI framework. – The current process of designing proteins in the lab is expensive and time-consuming, That’s why EvoDiff aims to simplify this process by generating high-quality proteins without requiring structural information about the target protein. – The framework has been trained on data from various species and functional classes of proteins.
Data analytics and AI software maker Databricks has raised over $500 million in a Series I funding round, increasing its valuation to $43 billion. – The round included investors such as T. Rowe Price, Morgan Stanley, Fidelity, Franklin Templeton, Capital One Ventures, Nvidia, Andreessen Horowitz, and Tiger Global.
Infosys has signed a $1.5 billion contract to leverage in AI solutions for a 15-year period. – The deal will involve Infosys providing enhanced digital experiences and business operation services using its platforms and AI solutions. The company has not disclosed the name of the client or whether it is an existing client.
Employees are feeling ‘AI Anxious’ — they’re worried about falling behind on AI and even pretending to be ‘in the know’ to colleagues
According to a new LinkedIn report, nearly half of the employees surveyed say they are worried they don’t know enough about AI. Nearly 40% said they pretend they know more about AI to seem ‘in the know’ in front of colleagues.
Survey Findings on AI Anxiety:
Concerns over AI familiarity: Nearly 49% of employees feel they should be more knowledgeable about AI.
Feeling of overwhelm: About 40% feel inundated with AI advancements and 56% are unsure about using AI at work.
Pretending knowledge: Approximately 40% admitted to feigning more AI knowledge than they possess to appear informed among peers.
Potential Job Disruptions and Shifts:
Fear of replacement: A study by Goldman Sachs highlighted potential disruption to over 300 million jobs globally due to generative AI.
Jobs at highest risk: Admin workers and women face the most significant risk of being replaced by AI.
Shift towards AI skills: A significant rise in job postings mentioning GPT or ChatGPT suggests a growing demand for AI skills.
Usage of AI Tools:
Unreported use: A Fishbowl survey found that 70% of 11,700 users used ChatGPT without informing their superiors.
Waning AI hype: Despite the growing concern, traffic to ChatGPT’s website has seen a decline for three consecutive months.
SpaceX’s Starlink satellite internet service registered $1.4 billion in 2022 revenue, significantly less than Elon Musk’s earlier projection of $12 billion.
Starlink reported its first profit in 2023, but the customer acquisition pace has been slower than expected, with one million active subscribers by 2022 end, well below Musk’s forecasted 20 million.
Starlink’s reliability was questioned by the FCC, which reversed its decision to grant Starlink $885.5 million funds, citing failure to meet program requirements.
Unity’s new pricing angers game developers
Unity Engine has scrapped its royalty-free licensing structure and will now charge developers on a “per-install” basis after certain thresholds are met.
The new Unity Runtime Fee, to take effect from January 1, 2024, will charge different per-install costs depending on the subscription tier.
This change has sparked outrage among the game development community who embraced Unity for its previous royalty-free structure.
Amazon unleashes AI for product descriptions
Amazon has launched an artificial intelligence tool for sellers to assist in creating compelling product page content.
The AI tool’s functions include generating product titles, descriptions, and listings, with the potential of enhancing seller success.
The drive to use AI in this way forms part of Amazon’s wider strategy to incorporate more AI applications into different aspects of its business.
EY’s $1.4 billion investment in AI and training
EY has announced the launch of EY.ai, a comprehensive platform aimed at facilitating clients’ AI adoption, in collaboration with major companies like Microsoft and IBM.
The firm has invested $1.4 billion into the platform, integrating AI into existing EY technologies such as EY Fabric, and is set to release a large language model called EY.ai EYQ.
EY’s CTO, Nicola Morini Bianzino, stated that the new proprietary AI offerings provide clients with confidence and a roadmap to successfully navigate their AI adoption journey.
Tesla engineers hated the Cybertruck so much they secretly designed an alternative
Many Tesla engineers were dissatisfied with the design of the company’s Cybertruck, leading them to secretly create alternative designs, according to chief designer Franz von Holzhausen.
Despite opposition, Elon Musk was adamant about maintaining the futuristic look of the Cybertruck, rejecting more traditional design concepts.
The Tesla Cybertruck sparked significant interest from public with over 200,000 orders made within the first three days of unveiling despite internal disagreements about its design.
SEC action against Mila Kunis and Ashton Kutcher’s NFT project
Mila Kunis and Ashton Kutcher’s NFT-based show, Stoner Cats, is charged by the SEC for selling unregistered securities, leading to a $1 million fine.
The original owners earned a 2.5% royalty every time a Stoner Cats NFT got resold, which the SEC claims is a return on investment.
A Fair Fund is established to return money to people who suffered financial loss from purchasing the NFTs, and Stoner Cats must destroy all NFTs they possess.
In Other Tech News on September 15th, 2023
Due to a micrometeoroid incident, NASA astronaut Frank Rubio extended his six-month space-stay to a year, setting a record for the longest continuous space duration by a NASA astronaut.
Despite initial blaming, Cruise robotaxis were not responsible for a man’s death in San Francisco; he was fatally hit by a city bus.
OpenAI plans to establish its first EU office in Dublin, hiring key legal, policy, and engineering roles, in preparation for impending regulatory challenges.
Google continues its wave of layoffs with significant cuts in the recruiting division, signaling an overall plan to reduce future hiring.
Cyber gang ‘Scattered Spider’ used phishing and malware to attack MGM Resorts, causing network paralysis and operational chaos for ransom payment.
Musk denies affair allegations with Brin’s wife, using a forced selfie, which Brin tried to avoid, as alleged in Musk’s new biography.
Emerging AI Innovations: September 14th, 2023
Can Large Language Models Reason?
This article by Melanie Mitchell explains the debate surrounding LLMs centers on whether they possess humanlike reasoning abilities or if their skills are merely a mirage. While LLMs have demonstrated emergent behaviors that resemble reasoning, it is unclear if these behaviors stem from true abstract reasoning or from memorizing and pattern-matching training data.
The implications of this debate are significant, as the presence of robust reasoning abilities in LLMs would support their role in developing trustworthy general intelligence. Conversely, if LLMs rely on memorization and pattern-matching, they will lack generalizability and cannot be trusted for tasks outside their training data.
Why does this matter?
This article hinges on whether they possess genuine human-like reasoning abilities or if their performance is primarily based on memorization and pattern-matching from training data. The outcome of this debate has substantial implications for the trustworthiness and generalizability of LLMs.
Microsoft Research’s new language model trains AI cheaper and faster
Microsoft Research has developed a new language model called phi-1.5 that could make training AI models cheaper and faster. The model uses curated synthetic data from existing large language models like OpenAI’s ChatGPT.
Despite having only 1 billion parameters compared to models with over 100 billion inputs, phi-1.5 has shown promising abilities with eliminating the need for web scraping or relying on data sources with copyright issues.
The model can reason and solve complex problems such as grade-school mathematics and basic coding. It exhibits traits of larger language models, both positive and negative, including the ability to think step by step and the potential for biased and toxic generations.
Why does this matter?
Despite having only 1 billion parameters compared to models with over 100B inputs, phi-1.5 has shown promising abilities and performs well on natural language tasks.
It eliminates the need for web scraping or relying on data sources with copyright issues. This research could democratize AI training by making it accessible to smaller organizations and reducing the reliance on specialized computing power.
Stability AI launches text-to-music AI
Stability AI has launched Stable Audio, a music and sound generation product. Stable Audio utilizes generative AI techniques to provide faster and higher-quality music and sound effects through a user-friendly web interface.
The product offers a free version for generating and downloading tracks up to 45 seconds long and a subscription-based ‘Pro’ version for commercial projects with 90-second downloadable tracks. Stable Audio allows users to input descriptive text prompts and desired audio length to generate customized tracks. The underlying model was trained using music and metadata from AudioSparx, a music library.
Why does this matter?
Stable Audio is the first music generation product that enables the creation of high-quality, 44.1 kHz music for commercial use via latent diffusion. It benefits end users by offering free and subscription-based options for customized music and sound effects, making audio creation accessible.
Potentially help empower content creators, businesses, and artists with an efficient tool to enhance their projects and creativity through tailored soundscapes.
Emerging AI Innovations: September 13th, 2023
NExT-GPT advances human-like AI research
The NExT-GPT system is a multimodal language model that can understand and generate content in various modalities, such as text, images, videos, and audio. It fills the gap in existing models by allowing for any multimodal understanding and generation.
NExT-GPT leverages pre-trained encoders and decoders, requiring only a small amount of parameter tuning. It also introduces a modality-switching instruction tuning (MosIT) and a curated dataset for complex cross-modal understanding.
Why does this matter?
This research demonstrates the potential for building AI agents to model universal modalities, advancing human-like AI research. It addresses the limitations in current models by enabling comprehensive multimodal understanding and generation, supporting seamless interactions and processing across various input formats.
Salesforce’s Einstein can customize AI for you
Salesforce introduced Einstein Copilot Studio, which allows customers to customize their AI offerings. The tool consists of three elements: prompt builder, skills builder, and model builder.
With the prompt builder, customers can add their own custom prompts for their products or brands.
The skills builder enables companies to add actions to prompts, such as competitor analysis or objection handling.
The model builder allows customers to bring their own models or use supported third-party offerings.
Salesforce is also working on a system called “the Einstein Trust Layer” to address issues like bias and inappropriate responses.
Why does this matter?
Customizing AI offerings will allow more personalized and tailored experiences to end users to align with their specific products or brands. This offering by Salesforce gives a competitive edge to its competitors by offering a comprehensive toolset for AI customization.
Apple silently making AI moves
Apple is quietly incorporating artificial intelligence into its new iPhones and watches to improve basic functions. The company showcased new gadgets with improved semiconductor designs that power AI features, such as better call quality and image capture.
Apple’s AI efforts have been reshaping its core software products behind the scenes without explicitly mentioning AI at its developer conference. Apple’s new watch chip includes a four-core “Neural Engine” that enhances Siri’s accuracy by 25% and enables new ways to interact with the device. The iPhone also automatically recognizes people in the frame for improved image capture.
Why does this matter?
Apple’s subtle integration of AI enhances the user experience and sets it apart from competitors like Microsoft and Google, who prioritize more ambitious AI transformations. These new AI additions can contribute to more efficient, intuitive, and enjoyable experiences for end users.
Amazon will pay an oil company to help it meet climate goals
Amazon has announced plans to purchase 250,000 metric tons of carbon removal from oil giant, Occidental Petroleum’s subsidiary 1PointFive, making Amazon the latest tech company to utilize carbon removal technologies to meet their climate goals.
1PointFive plans to build direct air capture (DAC) plants in Texas that pull carbon dioxide from the atmosphere and sequester it underground, a move being backed by both major tech companies and the Biden administration which has provided federal funding for similar projects.
Despite this new commitment, Amazon’s carbon removal purchase still represents only a fraction of the company’s emissions, with last year’s footprint amounting to more than 71 million metric tons of carbon dioxide.
More writers sue OpenAI
US novelist Michael Chabon and other writers have filed a proposed class action accusing OpenAI of copyright infringement for allegedly using their work to train AI models.
The lawsuit claims OpenAI sourced large amounts of content from across the internet, including from copyrighted works, to train its GPT models efficiently.
The writers allege that OpenAI’s AI generates in-depth analyses of themes present in their copyrighted works, indicating their works were used in the training process.
Daily AI Update News from Apple, Salesforce, Roblox, Alibaba, Coca-Cola, Character.ai, and Pika Labs
Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the day in the world of AI.
Apple is quietly incorporating AI into its new iPhones and watches to improve basic functions. – The company showcased new gadgets with improved semiconductor designs that power AI features, such as better call quality and image capture. Apple’s AI efforts have been reshaping its core software products behind the scenes, without explicitly mentioning AI at its developer conference. – Apple’s new watch chip includes a four-core “Neural Engine” that enhances Siri’s accuracy by 25% and enables new ways to interact with the device. The iPhone also automatically recognizes people in the frame for improved image capture.
Salesforce introduced Einstein Copilot Studio, allows customers to customize their AI offerings. – The tool consists of three elements: prompt builder, skills builder, and model builder. – With the prompt builder, customers can add their own custom prompts for their products or brand. – The skills builder enables companies to add actions to prompts, such as competitor analysis or objection handling. – The model builder allows customers to bring their own models or use supported third-party offerings. – Salesforce is also working on a system called “the Einstein Trust Layer” to address issues like bias and inappropriate responses.
Roblox, a popular online gaming platform, is set to introduce an AI chatbot called Roblox Assistant. – Which enables users to create virtual worlds simply by typing prompts. This development is particularly intriguing considering that 43% of Roblox users are under the age of 13.
Alibaba made its AI model ‘Tongyi Qianwen’, available to the public. – As it aims to compete with other Chinese tech giants like Baidu and Tencent. The company’s AI division, Alibaba Group Intelligence Group, has already collaborated with other firms to create applications based on Tongyi Qianwen. – The move comes as Alibaba’s new CEO, Eddie Wu Yongming, emphasizes the importance of AI and elevates it to one of the company’s main strategic focuses.
Coca-Cola has used AI to create a new flavor called Y3000 Zero Sugar. – The company collaborated with Bain & Company and OpenAI to develop the flavor and design the packaging using generative AI. Feedback from fans was taken into account during the formulation process. – Each can of Y3000 will feature a QR code that leads consumers to an online experience powered by AI, showcasing a vision of the year 3000.
Character.ai outperforms ChatGPT in mobile app usage within the US. – Character.ai, an AI app that allows users to create their own AI characters, is quickly catching up to ChatGPT in terms of mobile app usage in the US. – According to market intelligence firm Similarweb, Character.ai’s iOS and Android apps have gained 4.2 million monthly active users, just behind ChatGPT’s 6 million users. This highlights the growing popularity of Character.ai and its unique offering in the AI app market.
Pika Lab have Introduced Camera Movement Parameter. – It has control camera with customized intensity and direction and is available now for use.
How Tesla’s 36 Million ‘Eyes’ Drive the Future of Autonomy.
Tesla’s 36 Million ‘Eyes’ on the Streets.
The AI Data-Driven Path to Self-Driving.
Elon Musk showcased Tesla’s Full Self-Driving (FSD) technology on August 25. FSD12 uses a neural network planner, learning from billions of video frames to mimic human driving. Tesla’s extensive data and game-like motivation have led to impressive autonomous driving capabilities.
These eyes record continuously while the car is being driven, and in Sentry Mode when parked. The footage has been used to solve crimes, answer questions about collisions, and of course, train and improve the FSD. There is no other EV company gathering and harnessing as much data (from video and sensors) as Tesla and they count on around 10KH100 GPUs to do so.
Their FSD capabilities are far ahead and will likely remain so, especially with FSD12’s human-like teaching approach. If data is the new oil, Tesla is the oil king this is without mentioning their leadership in the EV charging station space among other sub-industries which allows them to gather more data and mine it with the help of AI to maximize profits and expand. Tesla’s focus on future tech and AI-driven data mining makes them a leader, not just in cars, but as an AI data juggernaut. While regulation challenges exist, the road to robotaxis and omnipresent FSD seems inevitable.
Note: Tesla hinted at the possibility of a 12 Alpha releaseby the end of the year*, it’s safer to assume that it could be 2024 before it’s widely available.*
Emerging AI Innovations: September 12th, 2023
AI technology has been making significant strides in recent years, but along with its advancements comes a concern for sustainability and its impact on the environment. One major aspect of this concern is the staggering water consumption rates of companies like Microsoft and Google. Last year alone, Microsoft’s data centers used an astounding amount of water, draining over 2,500 Olympic-sized swimming pools. This represents a 34% increase from the previous year. Similarly, Google reported a 20% increase in water consumption over the same period. The main reason behind this surge in water usage is the exponential growth of AI and its associated technologies. AI requires massive server farms, which heavily rely on water for cooling purposes. According to Shaolei Ren, a researcher at UC Riverside who focuses on AI’s environmental impact, the spike in water usage can be primarily attributed to AI. In fact, every 5 to 50 prompts submitted to ChatGPT, an AI language model, consumes approximately 500 ml of water. This information comes from an upcoming paper by Professor Ren’s team. Recognizing the gravity of the situation, tech giants like Google are taking responsible steps to address this issue. They have committed to assessing their water usage and its impact on surrounding areas. By doing so, they aim to find ways to mitigate any negative effects and promote responsible water usage. In conclusion, the rapid growth of AI technology has led to a concerning increase in water consumption by companies like Microsoft and Google. However, these tech companies are aware of the issue and are actively exploring ways to minimize their water usage and its environmental impact.
Google is taking a big step forward when it comes to responsible artificial intelligence (AI). They’ve just announced the launch of the Digital Futures Project and a whopping $20 million fund from Google.org. This fund will provide grants to top think tanks and academic institutions all over the world. So, what does this mean? Well, the project is aimed at supporting researchers, bringing people together for discussions, and encouraging the development of public policies that promote responsible AI. Now, let’s talk about some of the big names who have already received grants from the Digital Futures Fund. We’ve got the Aspen Institute, the Brookings Institution, the Carnegie Endowment for International Peace, the Center for a New American Security, the Institute for Security and Technology, SeedAI, and many more. But why does this really matter? Well, Google has always been at the forefront of advocating for responsible AI. They understand that perfecting AI is not something that any one company can do alone. It requires collaboration and input from academia and civil society. By supporting independent research across the globe, this move will help ensure that AI benefits everyone in a positive and transformative way. So, with Google’s Digital Futures Project and this impressive fund, we can look forward to a responsible future of AI.
So, let’s talk about the demand for Generative AI talent. It’s absolutely exploding right now, and the numbers prove it. Just take a look at these statistics. The number of companies mentioning “Generative AI” in their job postings each month is skyrocketing. It’s like a snowball rolling down a hill, getting bigger and bigger. And who’s leading the charge in hiring for GenAI talent? Well, it’s the big players like Amazon, Nvidia, Microsoft, and Google. But they’re not the only ones. Even big banks like Citigroup and CapitalOne are jumping on the GenAI train. It’s not surprising that the tech industry is the top sector in need of GenAI experts. They’re always at the forefront of innovation. Finance comes in at number two, and healthcare takes the third spot. But other sectors like real estate, basic materials, and energy aren’t showing as much interest. Now, here’s the interesting part. Companies are willing to pay big bucks for GenAI talent. Out of all the technical skills and technologies out there, jobs mentioning “Generative AI” or “LLMs” offer the highest average base salary. We’re talking about an average of $200,837 per year. Not too shabby, right? This all highlights the incredible impact that generative AI is having on various industries. It’s changing the game and businesses know it. And for job seekers and professionals, it means exciting opportunities. But it also means that you have to stay up to date with AI-related skills if you want to thrive in today’s market. So, yeah, Generative AI is kind of a big deal right now.
So, get this: researchers decided to challenge an AI-powered tech company to create a whopping 70 different programs in a study. And guess what? These chatbots managed to develop software in less than seven minutes! And hold onto your hats because it gets even crazier—they did it all for under a measly $1 in costs! Yep, you heard me right. The study put together this fictional software company called ChatDev, complete with stages like designing, coding, testing, and documenting. And get this—the AI workers actually collaborated at each stage. They made all sorts of decisions, from picking the right programming language to detecting bugs. Talk about teamwork! But it didn’t stop there. They even threw some real-world scenarios at ChatDev, like designing a Gomoku game. And what do you know? The results were pretty impressive. ChatDev managed to generate software systems with flawless performance about 86.66% of the time. Not too shabby, right? Now, don’t get me wrong. This AI-driven software development process still has its limitations. There’s a chance it could have some biases and such. But boy oh boy, does it show us what AI can do. It demonstrates the potential of automated software development with minimal human intervention. Imagine a world where AI runs a software company? It might just be closer than we think.
Meta is setting its sights on GPT-4, OpenAI’s powerful AI model, according to a recent report. They’re planning to train a new model that they hope will be just as impressive by investing heavily in data centers and H100 chips. Meta’s ambition is to create an AI model that surpasses their recent creation, Llama 2, by a significant margin. This new AI development is part of Meta’s strategy to reclaim its position as a prominent player in the AI field. They’ve formed a dedicated group, led by Mark Zuckerberg himself, which focuses on accelerating generative AI and creating tools that produce human-like expressions. The timeline for this project indicates that training for the AI system will commence in early 2024. Meta is not only investing in data centers but also acquiring advanced Nvidia chips (H100s) for their AI training. They’re shifting away from Microsoft’s cloud platform, Azure, and intend to train this new model on their own infrastructure. In a move towards openness, Zuckerberg plans to make the new AI model open-source. This approach allows companies to freely access and build AI-driven tools with it. Open-source AI models bring benefits like cost-effectiveness and flexibility, but they also carry risks, such as potential legal issues and misuse for spreading false information. Experts have expressed concerns about the new system’s unpredictability and potential vulnerabilities. They emphasize the importance of transparency and control to mitigate these risks.
Hey there! Guess what? Microsoft, MIT, and Google have been up to something really cool recently. They joined forces and created a research project called Large-Scale Automatic Audiobook Creation, and you won’t believe what they did with it! They managed to transform the ENTIRE Project Gutenberg Collection into audiobooks. Yep, you heard that right. The entire collection! Now, here’s the best part. These audiobooks are not just your average run-of-the-mill ones. They’re powered by AI, which means they’re not only free and open for everyone, but they sound pretty darn amazing too! The team used neural text-to-speech technology, and let me tell you, the quality of the voice acting is exceptional. You can even customize the speaking speed, style, and emotional intonation, and get this, you can even choose a specific voice for your audiobook with just a small sample of audio. How cool is that? But why is this such a big deal? Well, besides the fact that it’s super impressive, it also shows how text-to-speech AI can be used in practical ways. This project has created a scalable system that can convert thousands of e-books into high-quality audiobooks. That’s a huge leap forward for AI technology, and it’s exciting to see how it can solve real-world problems with a real impact.
Have you noticed the influx of ads for AI girlfriends and erotic chatbots on platforms like Instagram and TikTok? It seems like a whole new wave of bots has found a way to bypass moderation. It’s pretty fascinating how these sexually suggestive AI ads are popping up everywhere, despite the platforms banning many types of adult human content. These startups are cleverly using memes and animated characters to promote female avatars that promise adult experiences. And somehow, these ads slip through the cracks using sexually provocative messaging that’s similar to the banned content. It’s unclear why there’s looser enforcement when it comes to AI content. While some people might see this as harmless, others are concerned about the double standard that seems to be harming human sex workers. Commenters have called out problematic aspects, such as the appearance of minors in some of these ads. It’s definitely a difficult task to moderate AI content at such a large scale. But all of this is part of the rush to capitalize on the growing interest in generative AI. Developers are creating custom AI companions and experiences, and these explicit ads for AI chatbots and virtual companions are proliferating unchecked on social platforms. It’s quite a phenomenon, isn’t it?
Apple has just unveiled its latest lineup of iPhones, the iPhone 15 Pro and iPhone 15 Pro Max, and boy are there some exciting features to talk about! First up, let’s discuss the design. The new iPhones come with a sleek and durable titanium build, complete with contoured edges that not only look great but feel amazing in your hand. But that’s not all, folks! Apple has also introduced a new Action button, giving you even more control over your device. And let’s not forget about the camera upgrades. The base model of the iPhone 15 now boasts a 48-megapixel main sensor, allowing for incredible photo quality. Plus, it has optical zoom capabilities, giving you crisp and clear images even when zooming in. In terms of display, the iPhone 15 and iPhone 15 Plus feature 6.1-inch and 6.7-inch screens, just like last year. However, Apple has managed to take things up a notch by increasing the maximum brightness to a whopping 2000 nits. Talk about an immersive viewing experience! Now, let’s talk about colors. Apple has really stepped it up this year by embedding their color finishes throughout the glass, resulting in vibrant and eye-catching shades. You can choose from pink, yellow green, blue, and black, so there’s definitely something for everyone. And the best news? The iPhone 15 starts at the same price as last year, retailing at $799. You can even start preordering this Friday, with the official launch happening on September 22. So, whether you’re a mobile gamer, a photography enthusiast, or simply someone who wants a sleek and powerful device, the iPhone 15 Pro and iPhone 15 Pro Max are definitely worth considering. Don’t miss out on all the exciting features that Apple has packed into these beauties!
Hey there, it’s time for your daily dose of AI updates! Let’s jump straight into it. Google is making a big move to support responsible AI. They’re launching the Digital Futures Project and a $20 million Google.org fund. This fund will provide grants to think tanks and academic institutions around the world. With this project, Google aims to support researchers, organize discussions, and encourage the responsible development of AI by exploring public policy solutions. In exciting news, Microsoft, MIT, and Google have teamed up to transform the Project Gutenberg Collection into audiobooks. They’ve developed a system called Large-Scale Automatic Audiobook Creation, which can generate high-quality audiobooks from online e-books. Thanks to recent advancements in neural text-to-speech technology, the Project Gutenberg Open Audiobook Collection now offers thousands of free and open audiobooks. Who wouldn’t love to listen to a good book? Not to be left behind, Amazon, Nvidia, Microsoft, and Google are leading the charge in hiring GenAI talent. Generative AI is in great demand these days, with more and more companies mentioning it in their job postings. The technology sector is at the top of the list, followed by finance and healthcare. And guess what? Positions related to “Generative AI” or “LLMs” offer the highest average base salary of $200,837 per year. That’s definitely something to consider if you’re looking to dive into the field. Moving on to Instagram, it looks like we might soon be able to enjoy generative AI panoramas on the platform. This feature was spotted in a recent Instagram update for iOS. The company has been experimenting with various AI-related features, so the addition of AI-created panoramas doesn’t come as a huge surprise. It’s always exciting to see how AI can enhance our social media experience! In other news, the IRS is utilizing AI to crack down on tax violations among the wealthy. The agency is shifting its focus from working-class taxpayers to the wealthy, aiming to reduce the burden on average taxpayers. By leveraging AI and improved technology, they hope to identify sophisticated schemes used to avoid taxes. It’s a smart move to ensure everyone pays their fair share. YouTube is also getting in on the AI action. They’ve announced AI-powered creative guidance in Google Ads. This feature, available in the Recommendations and Video Analytics sections of Google Ads, provides suggestions to advertisers on how they can enhance the effectiveness of their video campaigns. AI is becoming an indispensable tool for optimizing marketing strategies. Now, here’s something mind-boggling. Researchers conducted a study where they tasked AI chatbots with running a tech company. And guess what? These chatbots were able to develop 70 different programs in under 7 minutes, costing less than $1. This demonstrates the power of AI in efficiently managing a software company with minimal human intervention. It’s a glimpse into the future of automated operations. Lastly, there’s a heartwarming story where ChatGPT, an AI chatbot, helped diagnose a boy’s chronic pain. The boy’s mother shared all the information she had about her son’s symptoms and his MRI scans with ChatGPT. And, lo and behold, the chatbot was able to identify the right diagnosis. This showcases the potential of AI to assist in healthcare and provide valuable insights. That’s all the AI updates for today! Stay tuned for more exciting developments in the world of artificial intelligence.
Google’s responsible AI leap
Google is launching the Digital Futures Project and a $20 million Google.org fund, which will provide grants to leading think tanks and academic institutions worldwide. The project will support researchers, organize convenings, and foster debate on public policy solutions to encourage the responsible development of AI.
Inaugural grantees of the Digital Futures Fund include the Aspen Institute, Brookings Institution, Carnegie Endowment for International Peace, the Center for a New American Security, the Institute for Security and Technology, SeedAI, and more. The fund will support institutions from countries around the globe.
Why does it matter?
Google has long been an advocate for responsible AI. But getting AI right will take more than any one company alone. This move will support many across academia and civil society to advance independent research on AI that helps this transformational technology benefit everyone.
Microsoft and Google’s staggering water consumption rates for AI
AI, with its vast resource needs, is raising concerns over sustainability and environmental impact. Last year, Microsoft’s data centers drained over 2,500 Olympic-sized swimming pools worth of water, reflecting a 34% increase from the previous year. Google also reported a 20% water consumption increase over the same period.
A deeper look at AI’s water footprint
The growth of AI and related technologies increases the need for vast server farms, which depend heavily on water for cooling purposes.
The spike in water usage can be attributed primarily to AI, as per Shaolei Ren, a researcher at the UC, Riverside, who focuses on AI’s environmental impact.
For every 5 to 50 prompts submitted to ChatGPT, it consumes about 500 ml of water, according to an upcoming paper from Professor Ren’s team.
Big Tech and responsible water usage
Recognizing their significant water consumption, tech companies like Google have voiced concerns and are exploring ways to mitigate the negative effects.
Google has committed to responsible water usage, which includes assessing where and how their water usage might affect surrounding areas.
Amazon, Nvidia, Microsoft, and Google lead hiring surge in GenAI
There is an explosive demand for Generative AI talent today. Here are some compelling statistics.
The number of companies mentioning “Generative AI” in monthly job postings is increasing exponentially.
Tech giants leading the surge in hiring for GenAI talent include Amazon, Nvidia, Oracle, Microsoft, Google, and more. Big banks like Citigroup and CapitalOne are also hiring big in GenAI.
Unsurprisingly, technology is the #1 sector looking to hire GenAI experts. Finance is #2nd, and healthcare is #3, while demand has been tepid in sectors like real estate, basic materials, and energy.
Companies are paying a lot for GenAI talent! Among all technical skills/technologies tracked, jobs mentioning “Generative AI” or “LLMs” had the highest average base salary offered, with an average of $200,837/year.
Why does this matter?
This reflects the pivotal role generative AI is playing across industries. Moreover, it signals a shift in how businesses are operating, adapting, and strategizing for an AI-led future. For job seekers and professionals, it presents exciting opportunities and emphasizes the need to stay updated with AI-related skills to thrive in the market today.
AI chatbots were tasked to run a tech company. They built software in under seven minutes — for less than $1
Researchers tasked an AI-powered tech company to develop 70 different programs in a new study. They found AI could develop software in under 7 minutes for less than $1 in costs, on average.
The ChatDev Experiment
Hypothetical Company: Researchers created a fictional software company, ChatDev, with stages including designing, coding, testing, and documenting.
Bot Collaboration: AI workers collaborated at each stage, making decisions like choosing a programming language and detecting bugs.
Real-world Simulation: ChatDev was tasked with real software development scenarios, like designing a Gomoku game.
Outcomes and Implications
Success Rate: About 86.66% of the software systems generated by ChatDev operated flawlessly.
Potential for the Future: While the system has its limitations, such as possible biases, it showcases the potential of AI-driven automated software development processes.
AI Efficiency: The experiment showcases the ability of AI like ChatGPT to operate a software company with minimal human intervention.
Meta sets GPT-4 as the bar for its next AI model, says a new report
Meta is reportedly planning to train a new model that it hopes will be as powerful as OpenAI’s GPT-4, by heavily investing in data centers and H100 chips. They hope the AI model will be way more powerful than Llama 2.
Meta’s AI Ambitions
New AI Development: Meta is working on an AI model, which they hope to be several times more powerful than their recent model, Llama 2.
Accelerating Generative AI: This initiative is spearheaded by a group established by Mark Zuckerberg earlier this year, focusing on AI tools that produce human-like expressions.
Expected Timeline: Meta anticipates the commencement of training for this AI system in early 2024.
Strategic Positioning in the AI Race
Behind Rivals: This new model is part of Zuckerberg’s strategy to reposition Meta as a leading entity in the AI domain after falling behind competitors.
Infrastructure Development: Meta is investing in data centers and acquiring advanced Nvidia chips (H100s) for AI training.
Shift from Microsoft: While Meta’s Llama 2 was integrated with Microsoft’s cloud platform, Azure, the new model is intended to be trained on Meta’s infrastructure.
Open-source Approach and Implications
Advocating Open-Source: Zuckerberg’s plan is to make the new AI model open-source, making it freely accessible for companies to build AI-driven tools.
Benefits and Risks: Open-source AI models are favored due to their cost-effectiveness and flexibility. However, they also come with potential downsides, including legal risks and misuse for disseminating false information.
Concerns from Experts: There are raised apprehensions about the unpredictability of the system and its potential vulnerabilities, emphasizing the need for transparency and control.
Microsoft, MIT, and Google transformed entire Project Gutenberg Collection into audiobooks
In a new research called Large-Scale Automatic Audiobook Creation, Microsoft, MIT, and Google collaborated to transform the entire Project Gutenberg Collection into audiobooks. The library now boasts thousands of free and open audiobooks powered by AI.
Utilizing recent advances in neural text-to-speech, the team achieved exceptional quality of voice acting. The system also allows users to customize an audiobook’s speaking speed and style, emotional intonation, and can even match a desired voice using a small amount of sample audio.
Why does it matter?
This presents an exceptional use case for text-to-speech AI. Moreover, it introduces a scalable system capable of converting thousands of HTML-based e-books to high-quality audiobooks. This signifies a remarkable leap in AI’s ability to solve real-world problems with tangible impact.
Ads for AI girlfriends are flooding Instagram and TikTok
Ads for AI girlfriends and erotic chatbots are bypassing moderation on platforms like Instagram and TikTok, a whole new wave of bots. (Source)
Sexually Suggestive AI Ads
Dozens of startups running NSFW ads on Meta and TikTok.
Use memes and animated characters to promote female avatars promising adult experiences.
Platforms have banned many types of adult human content.
But AI ads slip through using similar sexually provocative messaging.
Unclear why looser enforcement applied to AI content.
Critical Reactions
Some see double standard harming human sex workers.
Commenters call out problematic aspects like apparent minors.
But difficulty moderating AI content at scale persists.
Part of rush to capitalize on interest in generative AI.
Developers creating custom AI companions and experiences.
Explicit ads for AI chatbots and virtual companions are proliferating unchecked on social platforms, despite bans on similar sexual content from actual humans.
Apple unveils iPhone 15 Pro and iPhone 15 Pro Max: USB-C port, Dynamic Island, new colors
Summary: Featuring a strong and lightweight titanium design with new contoured edges, a new Action button, powerful camera upgrades, and A17 Pro for next-level performance and mobile gaming.The iPhone 15 stays at the same price as last year, starting at $799. Like last year, iPhone 15 and iPhone 15 Plus sport 6.1-inch and 6.7-inch displays. However, Apple said that maximum brightness can now go up to 2000 nits. You can preorder the iPhone 15 this Friday, with the device launching on September 22. Apple’s color finishes are now embedded throughout the glass, to form impressive new colorful finishes. The iPhone 15 is available in pink, yellow green, blue, and black. Apple has also updated the camera system. The base model iPhone 15 rear camera system now includes a 48-megapixel main sensor, like last year’s iPhone 14 Pro series. The 48-megapixel sensor enables 1x and 2x optical zoom modes, alongside the usual 0.5x ultra-wide lens.
Details: Apple today debuted iPhone 15 Pro and iPhone 15 Pro Max, designed with aerospace-grade titanium that’s strong yet lightweight to deliver Apple’s lightest Pro models ever. The new design also features contoured edges and a customizable Action button, allowing users to personalize their iPhone experience. Powerful camera upgrades enable the equivalent of seven pro lenses with incredible image quality, including a more advanced 48MP Main camera system that now supports the new super-high-resolution 24MP default, the next generation of portraits with Focus and Depth Control, improvements to Night mode and Smart HDR, and an all-new 5x Telephoto camera exclusively on iPhone 15 Pro Max. A17 Pro unlocks next-level gaming experiences and pro performance. The new USB‑C connector is supercharged with USB 3 speeds — up to 20x faster than USB 2 — and together with new video formats, enables powerful pro workflows that were not possible before.1 And with the addition of Roadside Assistance via satellite, the iPhone 15 Pro lineup builds on Apple’s innovative satellite infrastructure to connect users to help if they have car trouble while off the grid.
iPhone 15 Pro and iPhone 15 Pro Max will be available in four stunning new finishes, including black titanium, white titanium, blue titanium, and natural titanium. Pre-orders begin Friday, September 15, with availability beginning Friday, September 22.
“This is the most pro lineup we have ever created, with a state-of-the-art titanium design, the best iPhone camera system yet that enables game-changing new workflows, and the A17 Pro chip, which ushers in a new chapter of performance and games never before seen on iPhone,” said Greg Joswiak, Apple’s senior vice president of Worldwide Marketing. “iPhone 15 Pro and iPhone 15 Pro Max represent the best of Apple design and industry-first innovations to help enrich the everyday experiences of our users, while enabling them to unleash their creativity.”
A Stunning, Lightweight, and Durable Design
Available in 6.1-inch and 6.7-inch display sizes,2 iPhone 15 Pro and iPhone 15 Pro Max feature a strong and lightweight titanium design — a first for iPhone. This premium alloy — the same used in spacecraft — has one of the highest strength-to-weight ratios of any metal, making this Apple’s lightest Pro lineup ever. Both models feature a new refined brush texture, contoured edges, and the thinnest borders on iPhone. The Pro lineup is built to last, combining the strength of titanium with the toughest back glass in a smartphone and the industry-leading Ceramic Shield on the front. Using an industry-first thermo-mechanical process, the titanium bands encase a new substructure made from 100 percent recycled aluminum, bonding these two metals with incredible strength through solid-state diffusion. The aluminum frame helps with thermal dissipation and allows the back glass to be easily replaced. This new design highlights the Super Retina XDR display with Always-On and ProMotion technologies for an exceptional viewing experience.
The all-new Action button replaces the single-function switch used to toggle between ring and silent, offering additional options so users can choose between quickly accessing the camera or flashlight; activating Voice Memos, Focus modes, Translate,3 and accessibility features like Magnifier; or using Shortcuts for more options. A press-and-hold gesture with fine-tuned haptic feedback and visual cues in the Dynamic Island ensure the new button launches the intended action.
By default, the new Action button can switch between ring and silent, but users can choose from a set of actions for even more convenience and versatility.
A17 Pro: A New Generation of Apple Silicon for iPhone
Bringing pro performance and capabilities, iPhone 15 Pro and iPhone 15 Pro Max are powered by A17 Pro, the industry’s first 3-nanometer chip. Continuing Apple’s leadership in smartphone silicon, A17 Pro brings improvements to the entire chip, including the biggest GPU redesign in Apple’s history. The new CPU is up to 10 percent faster with microarchitectural and design improvements, and the Neural Engine is now up to 2x faster, powering features like autocorrect and Personal Voice in iOS 17. The pro-class GPU is up to 20 percent faster and unlocks entirely new experiences, featuring a new 6-core design that increases peak performance and energy efficiency. Now with hardware-accelerated ray tracing — which is 4x faster than software-based ray tracing — iPhone 15 Pro offers smoother graphics, as well as more immersive AR applications and gaming experiences. iPhone 15 Pro brings true-to-life gaming to the palm of users’ hands with console titles never before seen on a smartphone, like Resident Evil Village, Resident Evil 4, Death Stranding, and Assassin’s Creed Mirage.4
A17 Pro includes a dedicated AV1 decoder, enabling more efficient, high-quality video experiences for streaming services. Additionally, a new USB controller enables USB 3 speeds on iPhone for the first time, now supporting much higher transfer speeds and video output up to 4K at 60 fps HDR.
The new 6-core GPU in A17 Pro expands what’s possible on iPhone, enabling next-level mobile gaming with fast, efficient performance and hardware-accelerated ray tracing.
A Powerful Pro Camera System for More Creative Control
Through a deep integration of hardware and software, the advanced camera systems on iPhone 15 Pro and iPhone 15 Pro Max both pack the equivalent of seven pro lenses — all enabled by A17 Pro. With the power of computational photography, the 48MP Main camera, built exclusively for the Pro lineup, gives users even more flexibility with a new 24MP super-high-resolution default, offering incredible image quality at a practical file size ideal for storing and sharing. The Main camera allows users to switch between three popular focal lengths — 24 mm, 28 mm, and 35 mm — and even choose one as a new default. In addition to 48MP ProRAW, the Main camera also supports 48MP HEIF images with 4x more resolution. iPhone 15 Pro features an expansive 3x Telephoto camera, and iPhone 15 Pro Max provides the longest optical zoom ever on iPhone: 5x at 120 mm. Great for close-ups, wildlife photos, and catching the action from further distances, the new Telephoto camera on iPhone 15 Pro Max has an innovative tetraprism design with a combined optical image stabilization and autofocus 3D sensor-shift module, Apple’s most advanced stabilization system yet.
With Apple’s most pro camera systems ever, users will get powerful new features and the equivalent of seven camera lenses in their pocket, including 5x at 120 mm on iPhone 15 Pro Max.
Next-generation portraits on iPhone 15 Pro and iPhone 15 Pro Max feature sharper detail, more vivid colors, and better low-light performance. For the first time, users can take portraits without having to switch to Portrait mode. When there’s a person, dog, or cat in the frame, or when a user taps to focus, iPhone automatically captures depth information, so users can turn photos into stunning portraits later in the Photos app on iPhone, iPad, or Mac. For greater creative control, users can also adjust the focus point after the photo has been taken.
Additional features that benefit all cameras on iPhone 15 Pro and iPhone 15 Pro Max include:
Night mode gets better with sharper details and more vivid colors, now powered by the Photonic Engine, including Night mode portraits, enabled by the LiDAR scanner.
New Smart HDR captures subjects and the background with more true-to-life renderings of skin tones, while ensuring photos have brighter highlights, richer midtones, and deeper shadows when viewed in the Photos app. This advanced HDR rendering is also available to third-party apps, so images can look even better when shared online.
The best quality video in a smartphone is upgraded thanks to A17 Pro, with improvements in low-light video and Action mode.
The preferred smartphone for creative pros and filmmakers gets even better with new pro workflows. Users can now get up to 20x faster transfer speeds with an optional USB 3 cable. iPhone and third-party solutions like Capture One also help photographers create a pro studio, allowing them to shoot and instantly transfer 48MP ProRAW images from iPhone to Mac. ProRes video can be recorded directly to external storage, enabling higher recording options up to 4K at 60 fps, and greater flexibility on set when using iPhone as the main camera. iPhone 15 Pro also introduces a new option for Log encoding and is the first smartphone in the world to support ACES, the Academy Color Encoding System, a global standard for color workflows.
Coming later this year, iPhone 15 Pro will add a new dimension to video capture with the ability to record spatial video for Apple Vision Pro. Users will be able to capture precious moments in three dimensions and relive those memories with incredible depth on Apple Vision Pro when it is available early next year in the U.S.
Next-Level Wireless Performance and Connectivity
The iPhone 15 Pro lineup offers convenient new ways to charge, find friends in busy places, and stay connected while traveling. Both models use the USB‑C connector, a universally accepted standard for charging and transferring data, allowing the same cable to charge iPhone, Mac, iPad, and the updated AirPods Pro (2nd generation). Users can also charge AirPods or Apple Watch directly from iPhone with the USB‑C connector. iPhone 15 Pro and iPhone 15 Pro Max support USB 3 for data transfer speeds up to 10 gigabits per second, up to 20x faster than before.
Both models feature the second-generation Ultra Wideband chip, enabling two iPhone devices with this chip to connect at three times the range as before. This opens up a new way to use Precision Finding for Find My friends, so iPhone 15 users can share their location and find each other, even in crowds. Precision Finding is built with the same privacy protections that users have come to trust in Find My.5
iPhone 15 Pro and iPhone 15 Pro Max now support Wi-Fi 6E for greater wireless performance, including up to 2x faster speeds, and introduces the first Thread-enabled smartphones, opening up future opportunities for Home app integrations.
iPhone 15 Pro and iPhone 15 Pro Max come equipped with super-fast 5G,6 and include:
Support for MagSafe and future Qi2 wireless charging.
Improved audio quality on phone calls, including those made on FaceTime or third-party apps. Sound quality gets even better when users select Voice Isolation, so conversations come through loud and clear, even if they are somewhere noisy.
eSIM with support from more than 295 carriers. When traveling the world, users can stay connected through affordable international roaming plans from their existing carrier, or purchase prepaid eSIM plans in over 50 countries and regions, including Australia, Italy, Thailand, and more.
Expanded Safety Capabilities for Peace of Mind
The iPhone 15 lineup offers critical safety capabilities to provide assistance when it matters most, including Crash Detection7 and Emergency SOS via satellite.8 Currently available in 14 countries and regions on three continents, Emergency SOS via satellite has made a significant impact in users’ lives. This groundbreaking service will come to Spain and Switzerland later this month.
Building on this innovative satellite infrastructure, iPhone 15 Pro and iPhone 15 Pro Max introduce Roadside Assistance via satellite. Beginning in the U.S., when a user has car trouble and cellular and Wi-Fi coverage are not available, they can now connect to AAA, the country’s largest roadside assistance provider.9 An intuitive interface, including a short questionnaire to capture important details, will transmit the information via satellite so AAA can message with the user directly and dispatch help to their exact location. Access to Roadside Assistance via satellite will be included for free for two years. Service is covered according to AAA membership, but is also available separately for nonmembers.10
Featuring iOS 17
iPhone 15 Pro and iPhone 15 Pro Max feature iOS 17,11 making iPhone even more personal and intuitive with new features:
The Phone app gets major updates, with Contact Posters that allow users to customize how they appear to their contacts, and Live Voicemail, which leverages the power of A17 Pro to see real-time on-device transcription as someone leaves a voicemail. Users can even pick up the call while the caller is leaving their message.
Messages gets a new stickers experience, more powerful search, transcription of audio messages, and Check In, which allows users to automatically notify friends and family when they have made it to their destination safely.
NameDrop gives users a new way to use AirDrop to more easily share contact information by simply bringing two iPhone devices together. The same gesture can be used to AirDrop content and more, and now users can step away from each other and finish sending large files over the internet.12
StandBy gives users a customizable full-screen experience with glanceable information designed to be viewed from a distance when iPhone is on its side and charging. With the Always-On display of iPhone 15 Pro and iPhone 15 Pro Max, StandBy stays on to show useful information — perfect on a desk, nightstand, or kitchen counter.
Interactive widgets on the Home Screen, Lock Screen, and in StandBy allow users to take action with just a tap, making it easy to complete a to-do, or play or pause a song right from the widget.
Safari adds greater protection for Private Browsing and introduces profiles, helping users separate their browsing for topics like work and personal.
iOS 17 delivers many more updates, including Journal,13 a new app that helps iPhone users reflect and practice gratitude through journaling, improvements to autocorrect and Dictation that make entering text faster and easier than ever before, password and passkey sharing with iCloud Keychain, and much more.
Better for the Environment
iPhone 15 Pro and iPhone 15 Pro Max are designed with the environment in mind. As Apple continues to work toward its 2030 goal of making every product carbon neutral — from design to manufacturing to customer use — the company is prioritizing clean electricity across the entire supply chain and designing products with recycled and other low-carbon materials. iPhone 15 Pro and iPhone 15 Pro Max now use more recycled content, with a 100 percent recycled aluminum substructure and 100 percent recycled cobalt in the battery — both firsts for Apple. iPhone 15 Pro and iPhone 15 Pro Max also include 100 percent recycled rare earth elements in all magnets and 100 percent recycled gold in the USB‑C connector as well as the gold plating and tin soldering in multiple printed circuit boards. Both models meet Apple’s high standards for energy efficiency and are free of mercury, PVC, and beryllium. Over 99 percent of the packaging is fiber-based, bringing Apple closer to its goal of completely removing plastic from its packaging by 2025.
To further reduce impact on the planet, Apple will no longer use leather in any new Apple products, including iPhone accessories. Apple is introducing a new FineWoven Case with MagSafe and FineWoven Wallet with MagSafe, made from a durable and elegant microtwill with a soft, suedelike feel. The material is made from 68 percent post-consumer recycled content and has significantly lower carbon emissions compared to leather.
Pricing and Availability
iPhone 15 Pro and iPhone 15 Pro Max will be available in black titanium, white titanium, blue titanium, and natural titanium finishes. iPhone 15 Pro remains at the same starting price of $999 (U.S.)or $41.62 (U.S.) per month, available in 128GB, 256GB, 512GB, and 1TB storage capacities. iPhone 15 Pro Max starts at $1,199 (U.S.) or $49.95 (U.S.)per month, available in 256GB, 512GB, and 1TB storage capacities.
Apple offers great ways to save and upgrade to the latest iPhone. Customers in the U.S. can get $200–$650 (U.S.) in credit when they trade in an iPhone 11 or later and upgrade to iPhone 15 Pro or iPhone 15 Pro Max by visiting the Apple Store Online, or at an Apple Store location. To see what their device is worth and for terms and conditions, customers can visit apple.com/shop/trade-in.
Customers can get iPhone 15 Pro for as low as $0 (U.S.) after a qualifying trade-in with select U.S. carriers. For eligibility requirements and more details, see apple.com/shop/buy-iphone/carrier-offers.
Customers in more than 40 countries and regions, including Australia, Canada, China, France, Germany, India, Japan, Mexico, the UAE, the U.K., and the U.S., will be able to pre-order iPhone 15 Pro and iPhone 15 Pro Max beginning at 5 a.m. PDT this Friday, September 15, with availability beginning Friday, September 22.
iPhone 15 Pro and iPhone 15 Pro Max will be available in Macao, Malaysia, Türkiye, Vietnam, and 17 other countries and regions beginning Friday, September 29.
FineWoven Wallet with MagSafe and FineWoven Case with MagSafe will both be available for $59 (U.S.) in five new colors for the iPhone 15 lineup: black, taupe, mulberry, pacific blue, and evergreen. In addition to the iPhone 15 Pro and iPhone 15 Pro Max Clear Case, available for $49 (U.S.), a Silicone Case with MagSafe will be available for $49 (U.S.) in black, storm blue, clay, light pink, guava, orange sorbet, cypress, and winter blue.
iOS 17 will be available as a free software update on Monday, September 18.
Beginning September 18, iCloud+ will offer two new plans: 6TB for $29.99 (U.S.) per month and 12TB for $59.99 (U.S.) per month, providing additional storage to keep files, photos, videos, and more safe, accessible, and easy to share. The new plans are great for users with large photo and video libraries or those using Family Sharing, and will provide access to premium features, including Private Relay, Hide My Email, Custom Email Domains, and HomeKit Secure Video support.
Customers who purchase iPhone 15 Pro and iPhone 15 Pro Max will receive three free months of Apple Arcade and Apple Fitness+ with a new subscription.
Daily AI Update News from Google, Microsoft, Instagram, YouTube, and more.
Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the day in the world of AI.
Google’s new leap to support responsible AI – It is launching the Digital Futures Project and a $20 million Google.org fund, which will provide grants to leading think tanks and academic institutions worldwide. The project will support researchers, organize convenings, and foster debate on public policy solutions to encourage the responsible development of AI.
Microsoft, MIT, and Google transformed the entire Project Gutenberg Collection into audiobooks – In new research called Large-Scale Automatic Audiobook Creation, the institutes present a system that can automatically generate high-quality audiobooks from online e-books. The Project Gutenberg Open Audiobook Collection now boasts thousands of free and open audiobooks powered by AI. Utilizing recent advances in neural text-to-speech, the team achieved exceptional quality of voice acting.
Amazon, Nvidia, Microsoft, and Google are leading the surge in hiring for GenAI talent – There is an explosive demand for Generative AI talent today. The number of companies mentioning “Generative AI” in monthly job postings is increasing exponentially. Technology is the #1 sector looking to hire GenAI experts. Finance is #2nd, and healthcare is #3. AI jobs mentioning “Generative AI” or “LLMs” had the highest average base salary offered, with an average of $200,837/year.
Instagram might be getting generative AI panoramas – The feature was spotted in Monday’s Instagram update on iOS. It has been experimenting with a handful of generative AI-related features, so it’s not too surprising that AI-created panoramas might be part of Instagram’s feature lineup.
IRS deploys AI to crack down on tax violations among nation’s wealthy – The agency focus will shift attention to wealthy from working-class taxpayers; key changes coming to reduce burden on average taxpayers while using AI and improved technology to identify sophisticated schemes to avoid taxes.
YouTube announces AI-powered creative guidance in Google Ads – The AI-powered feature, found inside the Recommendations and Video Analytics sections of Google Ads, offers suggestions to help advertisers enhance the efficacy of their video campaigns.
AI chatbots were tasked to run a tech company. They built software in under 7 minutes for less than $1 – Researchers in this new study tasked an AI-powered tech company with developing 70 different programs. AI chatbots such as ChatGPT can operate a software company in a quick, cost-effective manner with minimal human intervention, the study indicates.
A boy saw 17 doctors over 3 years for chronic pain. ChatGPT found the right diagnosis – The frustrated mom of the boy shared with ChatGPT everything she knew about her son’s symptoms and all the information she could gather from his MRIs.
Emerging AI Innovations: September 11th, 2023
Google Deepmind introduces language models as optimizers
Google DeepMind introduces the concept of using language models as optimizers, This work is called Optimization by PROmpting (OPRO). This new approach describes the optimization problem in natural language. The models are trained to generate new solutions based on a defined problem and previously found solutions.
This is applied to linear regression, traveling salesman problems, and prompt optimization tasks. The results show that the prompts optimized by OPRO outperform human-designed prompts by up to 8% on GSM8K and up to 50% on Big-Bench Hard tasks.
Why does this matter?
Google Deepmind’s OPRO can revolutionize problem-solving in various fields. It improves task accuracy, outperforming human-designed approaches benefiting end users with more efficient solutions.
NVIDIA’s new software boosts LLM performance by 8x
NVIDIA has developed a software called TensorRT-LLM to supercharge LLM inference on H100 GPUs. It includes optimized kernels, pre- and post-processing steps, and multi-GPU/multi-node communication primitives for high performance. It allows developers to experiment with new LLMs without deep knowledge of C++ or NVIDIA CUDA. The software also offers an open-source modular Python API for easy customization and extensibility.
(The following figures reflect performance comparisons between an NVIDIA A100 and NVIDIA H100.)
Additionally, it allows users to quantize models to FP8 format for better memory utilization. TensorRT-LLM aims to boost LLM deployment performance and is available in early access, soon to be integrated into the NVIDIA NeMo framework. Users can apply for access through the NVIDIA Developer Program, with a focus on enterprise-grade AI applications.
Why does this matter?
H100 alone is 4x faster than A100. Adding TensorRT-LLM and its benefits, including in-flight batching, results in an 8X total increase to deliver the highest throughput. Also, on Meta’s Llama 2 TensorRT-LLM can accelerate inference performance by 4.6x compared to A100 GPUs.
Companies like Databricks have found TensorRT-LLM to be easy to use, feature-packed, and efficient, enabling cost savings for customers.
Google’s antitrust trial to begin
The ongoing trial against Google is part of a larger reassessment of internet regulation, with antitrust enforcers fearing large tech companies could stifle innovation in developing fields such as AI.
The Justice Department, employing the Sherman Antitrust Act of 1890, continues its case against Google’s market dominance, implying penalties could range up to the company’s breakup.
Measures to rein in industry giants, however, may dampen innovation, signaling caution for regulators as global competition, notably with China, intensifies.
Silicon Valley’s pursuit of immortality
Silicon Valley elites are embracing a trend of costly full-body MRIs, like those offered by Prenuvo, as a preventative health measure to detect diseases early.
Despite their rising popularity among the wealthy and endorsements by celebrities, there’s skepticism in the medical community regarding the scans’ effectiveness and potential for false positives.
Start-ups such as Ezra and Prenuvo aim to reduce scan costs through AI technology and seek to bring these preventative measures into the mainstream.
Potential world’s largest lithium cache discovered in the US
A new study suggests that the McDermitt Caldera, located on the Nevada-Oregon border, may contain the world’s largest lithium deposit, estimated between 20 to 40 million metric tons.
This lithium, vital for creating batteries that power electric vehicles, is trapped within clay, specifically within an area called Thacker Pass located in southern Nevada.
Although this colossal lithium deposit could reshape global lithium dynamics, its extraction has faced opposition due to concerns about its environmental impact and disruption of sacred Native American lands.
Daily AI Update News from NVIDIA, Google DeepMind, Meta, Reddit, Infosys, India’s Reliance and biased GPT model
Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the day in the world of AI.
NVIDIA has developed TensorRT-LLM to supercharge LLM inference on H100 GPUs – It includes optimized kernels, pre- and post-processing steps, and multi-GPU/multi-node communication primitives for high performance. – It allows developers to experiment with new LLMs without deep knowledge of C++ or NVIDIA CUDA. – The software also offers an open-source modular Python API for easy customization and extensibility. – Companies like Databricks have found TensorRT-LLM to be easy to use, feature-packed, and efficient, enabling cost savings for customers.
Google DeepMind introduces the concept of using language models as optimizers – This new approach describes the optimization problem in natural language. The models are trained to generate new solutions based on a defined problem and previously found solutions. – This approach was tested on linear regression and the traveling salesman problem, and the results showed that the language models performed as well as or better than hand-designed algorithms. – The idea was then applied to prompt optimization for maximizing task accuracy, such as math word problem-solving. The optimized prompts outperformed human-designed prompts by over 50% in some cases.
Meta training a new model with aim to match GPT-4 level AI model as its benchmark, says a new report – Meta is reportedly planning to train a new chatbot model that it hopes will rival OpenAI’s GPT-4. The company is acquiring AI training chips and expanding its data centers to create a more powerful chatbot. – CEO Mark Zuckerberg wants the model to be free for companies to create AI tools with. Meta is building the model to speed up the creation of AI tools that can emulate human expressions.
Reddit launched AI-powered keyword research tool, that will help advertisers – This tool uses ML and NLP to generate relevant keywords and rank them based on monthly Reddit views. – It filters out unsuitable content to ensure brand safety. Advertisers can use this tool to display the most fitting ads to the most relevant Reddit users, saving time on keyword research and increasing the reach and efficiency of campaigns.
Infosys likely to collab with NVIDIA to train 3 lakh+ employees on AI – NVIDIA CEO Jensen Huang hinted at the collaboration, stating that Infosys would use NVIDIA’s infrastructure and capabilities to build AI models and applications. – Infosys has already launched an AI-focused suite called Topaz, which offers industry-specific solutions in intelligent automation and enhanced security.
India’s Reliance partners with Nvidia to develop a new LLM – Reliance Industries’ Jio Platforms has partnered with Nvidia to build a LLM trained on India’s diverse languages. They will also collaborate on building an AI infrastructure that is more powerful than India’s fastest supercomputer. – Reliance will manage the AI cloud infrastructure, while Nvidia will provide AI supercomputer solutions and frameworks. India has yet to make a significant mark in the global AI arena, with most companies relying on LLM created by organizations like OpenAI.
Researchers at Humboldt University in Berlin have developed a very biased GPT model, called OpinionGPT – A language model that demonstrates how training data can impact AI bias. The model is trained on specific social dimensions such as politics, geography, gender, and age using data from selected thematic subgroups on Reddit. – While the model does not represent all demographics, it reflects the nuanced biases of different demographics. This research highlights the importance of carefully selecting and curating training data to mitigate bias in AI models. It is called a very biased GPT model.
Emerging AI Innovations: September 08th, 2023
Introducing Falcon 180B, largest and most powerful open LLM
UAE’s Technology Innovation Institute (TII) has released Falcon 180B, a new state-of-the-art for open models. It is the largest openly available language model, with 180 billion parameters, trained on a massive 3.5 trillion tokens using TII’s RefinedWeb dataset. It’s currently at the top of the Hugging Face Leaderboard for pre-trained Open LLMs and is available for both research and commercial use.
The model performs exceptionally well in various tasks like reasoning, coding, proficiency, and knowledge tests, even beating competitors like Meta’s LLaMA 2. Among closed-source models, it ranks just behind OpenAI’s GPT 4 and performs on par with Google’s PaLM 2 Large, which powers Bard, despite being half the model’s size.
Why does it matter?
It is a great contribution to open source. But there’s a catch: You’ll need 400GB of memory for inference, which can cost too much to host this for inference. Moreover, code is only 5% in the training mix, which is by far the most useful data to boost reasoning, master tool use, and power AI agents.
However, this indicates the continuous pushing of the boundaries of generative AI, and we may be only a few months away from GPT-4-level open-source models.
Apple is spending millions of dollars a day to train AI
Reportedly, Apple has been expanding its budget for building AI to millions of dollars a day. It has a unit of around 16 members, including several former Google engineers, working on conversational AI. It is working on multiple AI models to serve a variety of purposes.
Apple wants to enhance Siri to be your ultimate digital assistant, doing multi-step tasks without you lifting a finger and using voice commands.
It is developing an image generation model and is researching multimodal AI, which can recognize and produce images or video as well as text.
A chatbot is in the works that would interact with customers who use AppleCare.
Why does it matter?
OpenAI, too, splashed out a whopping $100 million for GPT-4 alone. Perhaps this shouldn’t be surprising, given that Apple has been a visionary in the past, consistently pushing the boundaries of what’s possible in technology. It is also reported that Apple created a team four years ago, indicating it may not be as much of a laggard in the AI race as we thought.
Microsoft and Paige to build the largest image-based AI model to fight cancer
Paige, a technology disruptor in healthcare, has joined forces with Microsoft to build the world’s largest image-based AI models for digital pathology and oncology.
Paige developed the first Large Foundation Model using over one billion images from half a million pathology slides across multiple cancer types. Now, it is developing a new AI model with Microsoft that is orders-of-magnitude larger than any other image-based AI model existing today, configured with billions of parameters.
Paige will utilize Microsoft’s advanced supercomputing infrastructure to train the technology at scale and ultimately deploy it to hospitals and laboratories across the globe using Azure.
Why does this matter?
This will help realize the potential of generative AI at an unprecedented scale, introduce completely novel capabilities of AI, and serve as the cornerstone for the next generation of clinical/healthcare applications built with AI.
Vector Database: The Secret Behind Large Language Models Capabilities
Have you ever wondered how language models like GPT-3, BERT, and others seem to understand and generate text with astonishing accuracy? The answer lies in vector embeddings.
Vector Embeddings Explained
Vector embedding is a prominent method in artificial intelligence and natural language processing that transforms data into lower-dimensional vectors. These embeddings encapsulate the crux of information, allowing AI systems to delve deeper into the data and enhance memory retention. Think of embeddings as a streamlined bridge between raw data and an AI’s comprehension capability.
AI models, like LLMs, generate embeddings that capture intricate relationships and hidden structures by converting data into these vectors. As a result, the AI can discern patterns and relationships more effectively.
Why Vector Databases Are Essential
Traditional scalar databases falter when faced with the multifaceted nature of vector embeddings. These databases, crafted for simpler, one-dimensional data, struggle with the elevated complexity of multi-dimensional vectors. Here’s why vector databases are the answer:
Complexity and Scale: The richness of vector embeddings can overwhelm conventional databases, making querying and processing a challenge.
Gaining Insights: To glean substantial insights from vector data, specialized techniques are needed, which many traditional databases lack.
Real-time Analysis: AI operations often demand instantaneous analysis. However, the demands of vector embeddings can bog down traditional databases, limiting their real-time response.
Vector Databases in Focus: Specifically built for vector data, these databases offer optimized tools for storage, indexing, and querying. With indexing techniques tailored for high-dimensional spaces, they ensure rapid data retrieval.
Peak Performance: Vector databases harness hardware capabilities, such as GPUs and TPUs, ensuring optimal performance even with voluminous vector data.
Scalability: Unlike traditional databases that may flounder with growing data, vector databases scale efficiently, ensuring undiminished performance.
Adaptability: These databases permit varied data representations and querying methods, giving users the freedom to experiment without any database limitations.
In essence, to fully harness the potential of vector embeddings, vector databases are the go-to. They pave the way for quicker insights, on-the-spot analysis, and smarter decision-making in AI.
Understanding Vector Databases with an Example
Let’s demystify the workings of a vector database with a real-world example: chatbots, such as chatGPT or bard, which are grounded in LLMs. Here’s a step-by-step breakdown:
A user keys in a query into the chatbot.
This query is processed by an embedding model to produce a relevant vector embedding.
The generated vector embedding is then indexed in the vector database, linking it to the original content.
Upon querying, the vector database swiftly retrieves the most relevant response and presents it to the user.
In this manner, vector databases power chatbots, offering quick and pertinent responses to user queries.
Generate innovative business ideas with ChatGPT
Brainstorming innovative business ideas requires a blend of human creativity and machine intelligence. With ChatGPT, you can harness the latter to get a fresh perspective on business opportunities.
Use this prompt to explore how to utilize ChatGPT to generate actionable and unique business concepts.
I want you to act as a world-class business strategist specializing in innovation and entrepreneurship. My first request is for you to generate a business idea:
Here's some context:
Industry — [Industry]
Interest — [Interests]
Formatting guidelines: "Include: Problem Statement, Solution, Unique Selling Point (USP), Target Market, and Revenue Model. Include some context and a list of [Number] possible business ideas.".
I need you to write content with a good balance of “perplexity” and “burstiness”.
Daily AI Update News from TII, Apple, Microsoft, OpenAI, IBM, Anthropic, Slack, HubSpot, and SAP
OpenAI to host its first developer conference in November – The one-day event in San Francisco will unite developers from around the world with the team at OpenAI to preview new tools, exchange ideas, and join breakout sessions led by OpenAI’s technical team members. Registration for in-person attendance will open soon, and developers everywhere can livestream the keynote.
IBM rolls out new generative AI features and models – It includes new capabilities across its recently launched Watsonx data science platform and new models called the Granite series models, appearing to be standard LLMs along the lines of GPT-4 and ChatGPT.
Anthropic launches a paid plan for Claude 2 – Claude Pro is priced the same as OpenAI’s ChatGPT Plus, the paid plan for Claude 2 rival ChatGPT. For the monthly $20 in the U.S., or £18 in the U.K., customers get 5x more usage than the free Claude 2 tier, the ability to send “many more” messages, priority access to Claude 2 during high-traffic periods, and early access to new features.
Slack launches new Workflow Builder to help better automate your tasks – Aimed at users with little or no coding experience, the new version of Workflow Builder tries to connect different tools to automate your workflows.
HubSpot announced the launch of HubSpot AI – It is a platform-wide portfolio of AI-powered features for marketing, sales, and service teams. It includes AI Assistants, AI Agents, AI Insights, and ChatSpot.
SAP acquires LeanIX to focus on AI-assisted IT modernization – LeanIX, a German startup, provides enterprises with a clear picture of their entire software usage, covering everything from what they’ve bought, licensed, and built to what they plan to add. The move will accelerate modernization for enterprise customers.
Emerging AI Innovations: September 07th, 2023
Harvard Scientist Dr. Isaac Kohane Validates GPT-4’s Medical Prowess with Rare Diagnosis
Dr. Isaac Kohane, a dual expert in medicine and computer science at Harvard, recently evaluated the capabilities of the latest AI model, GPT-4, in a clinical context. Remarkably, GPT-4 showcased proficiency surpassing many medical professionals, achieving a 90% accuracy rate on medical licensing exam questions, providing patient translations, and offering insights into physician-patient interactions.
During a real-world test, GPT-4 astoundingly identified a rare medical condition, matching the diagnostic skills of seasoned physicians like Kohane. Yet, his recent publication, ‘The AI Revolution in Medicine,’ sheds light on the model’s occasional lapses, from minor clerical oversights to numerical inaccuracies.
Amidst the marvel and skepticism, Kohane grapples with the challenge of ensuring the AI’s consistent reliability in medical consultations. Read the full article here.
Yokosuka Leads Japanese Municipalities with Innovative AI Implementation in Public Offices
In a pioneering initiative, Yokosuka emerges as the first city in Japan to integrate OpenAI’s ChatGPT into its municipal offices. This one-month experiment, encompassing 4,000 personnel, is designed to refine administrative procedures. Amid a diminishing populace and staffing constraints, the city aims to utilize AI-powered solutions to optimize manpower for endeavors demanding a personalized touch.
ChatGPT is envisioned to support in roles such as content summarization, brainstorming marketing text, formulating official documents, and refining comprehensible language. This strategic decision aligns with the broader Japanese governmental agenda, keenly assessing AI’s potential role in streamlining state-run administrative operations. Read the detailed report here.
OpenAI doubles GPT-4 message cap to 50
OpenAI has doubled the number of messages ChatGPT Plus subscribers can send to GPT-4. Users can now send up to 50 messages in 3 hours, compared to the previous limit of 25 messages in 2 hours. And they are rolling out this update next week. Read the detailed report here.
GPT-4 Code Interpreter masters math with self-verification
OpenAI’s GPT-4 Code Interpreter has shown remarkable performance on challenging math datasets. This is largely attributed to its step-by-step code generation and dynamic solution refinement based on code execution outcomes.
Expanding on this understanding, new research has introduced the innovative explicit code-based self-verification (CSV) prompt, which leverages GPT4-Code’s advanced code generation mechanism. This prompt guides the model to verify the answer and then reevaluate its solution with code.
The approach achieves an impressive accuracy of 84.32% on the MATH dataset, significantly outperforming the base GPT4-Code and previous state-of-the-art methods.
ChatGPT will now remember who you are & what you want
OpenAI is rolling out custom instructions to give you more control over how ChatGPT responds. It allows you to add preferences or requirements that you’d like ChatGPT to consider when generating its responses.
ChatGPT will remember and consider the instructions every time it responds in the future, so you won’t have to repeat your preferences or information. Currently available in beta in the Plus plan, the feature will expand to all users in the coming weeks.
OpenAI Unveils Shap·E: A Revolutionary AI for Rapid 3D Design Generation
OpenAI introduces its innovative creation, Shap·E – a cutting-edge conditional generative model primed for swift generation of 3D designs. Engineered using implicit functions, it can seamlessly render as textured meshes or neural radiance landscapes.
Rooted in a robust dataset amalgamating 3D assets with matching textual narratives, Shap·E employs an encoder to translate these assets into implicit function parameters. Furthermore, a conditional diffusion model refines the conditional distribution of these parameters, based on input data. The result? A transformative AI tool that can manifest intricate, high-caliber 3D outputs in mere moments. Discover the complete story here.
ChatGPT in your pocket: OpenAI brings ChatGPT to iPhone
OpenAI has released the ChatGPT app for iOS that allows users to have engaging and personalized conversations with an AI language model. It will sync your conversations, support voice input, and bring the latest model improvements to your fingertips. Also, ChatGPT Plus subscribers get exclusive access to GPT-4’s capabilities, early access to features, and faster response times, all on iOS.
They have started the rollout in the US and are expected to enter additional countries in the coming weeks. OpenAI said: “Android users, you’re next!”. Discover the complete story here.
OpenAI Announces $1M Initiative for Democratizing AI Governance
In a groundbreaking move, OpenAI pledges to distribute ten grants, each worth $100,000, to pioneering projects focused on formulating democratic processes to steer AI system regulations and actions.
Acknowledging AI’s transformative influence on global society, OpenAI calls upon visionaries to conceptualize unique systems that imbibe an array of viewpoints and are anchored in the collective good. This initiative primarily seeks to unravel questions surrounding AI behaviors, delving into scenarios like the grounds on which AI might critique public personalities or the manner in which it portrays contested perspectives.
Prospective participants have until June 24, 2023, to present their ideas, with the flexibility to either pick from a curated list of policy dilemmas or introduce their original thoughts. Dive deeper into the initiative here.
OpenAI’s massive update on GPT-3.5 & GPT-4 APIs
OpenAI announced exciting updates, including more steerable API models, function calling capabilities, longer context, and lower prices.
Function calling is now available to enable LLMs to work more effectively and efficiently interact with your programs/tools.
The latest GPT-3.5-Turbo model is available with support for 16K context (~20 pages of text)
25% cost reduction on input tokens for gpt-3.5-turbo
75% cost reduction on the embeddings model
The data privacy and security assurances implemented on March 1 remain consistent across all models. The user’s API data will not be utilized for training purposes. Dive deeper into the initiative here
Generative AI poised to replace 2.4 million US jobs by 2030
Forrester predicts that generative AI will replace 2.4 million US jobs by 2030, mostly white-collar roles, such as technical writers, proofreaders, copywriters, and administrative positions. But ironically, other forms of automation will displace more jobs.
To stay on top of the latest advancements in AI, look here first.
Concerns about Generative AI
While the Generative AI impact is significant, other forms of automation are set to cause more widespread job displacement.
The most impacted group will be middle-class, college-educated, white-collar workers, specifically those earning above $60,000 annually.
Creative professionals stand to benefit
Interestingly, workers in creative industries will likely utilize generative AI tools in their jobs rather than being replaced. This includes editors, writers, authors, poets, and lyricists.
However, the use of such tools like ChatGPT may result in inconsistent outputs and even “coherent nonsense”, leading to potential performance issues.
Microsoft Research presents an experimental study using OpenAI’s ChatGPT for robotics applications. It outlines a strategy that combines design principles for prompt engineering and the creation of a high-level function library that allows ChatGPT to adapt to different robotics tasks, simulators, and form factors.
The study encompasses a range of tasks within the robotics domain, from basic logical, geometrical, and mathematical reasoning to complex domains such as aerial navigation, manipulation, and embodied agents.
Microsoft also released PromptCraft, an open-source platform where anyone can share examples of good prompting schemes for robotics applications.
Connected cars are a “privacy nightmare,” Mozilla Foundation says
Modern cars collect enormous amounts of personal data and many car manufacturers insist on sharing or selling this sensitive information, such as disability status, genetic information and facial templates, according to a comprehensive study by the Mozilla Foundation.
The vast majority (92%) of car companies give drivers little to no control of their personal data, with the notable exceptions of Renault and Dacia, both European brands obligated to comply with GDPR privacy laws.
Nissan’s privacy policy stands out as the most concerning, bluntly stating they can collect and share highly sensitive personal information for targeted marketing, but privacy concerns extend across the board, with all companies collecting substantial driver data, evidencing a broader problem with privacy in the car industry.
ChatGPT’s biggest competitor launches Pro version at $20
Anthropic introduced a paid Claude Pro plan for its Claude 2 chatbot, offering 5x more usage than the free tier to compete with OpenAI’s ChatGPT Plus. (Source)
The Paid Offering
$20 monthly subscription in the U.S.
5x more usage than free Claude 2.
Priority access during high traffic.
Early new feature access.
Motivations
Matches price of rival ChatGPT Plus.
Revenue helps fund costly compute for models.
Limits still in place due to compute constraints.
Seeking billions more to develop envisioned AI assistant.
Anthropic launched a $20 Claude Pro subscription offering 5x more usage of its Claude 2 AI assistant, seeking revenue to fund its bid to compete with rivals like OpenAI.
Elon Musk Plans to Merge Neuralink and Tesla for an AI Supercompany
Elon Musk reportedly plans to blend Neuralink and Tesla into a large AI company, using data from Twitter users and Tesla’s Full Self-Driving Cameras to train a robust AI model.
Musk’s AI Integration Plan
Musk is contemplating merging Neuralink and Tesla, alongside his xAI startup, to create a comprehensive artificial intelligence model.
Leveraging the text data from Twitter and real-world images from Tesla’s Full Self-Driving network, he intends to develop AI chatbots and physical robots capable of real-world navigation.
Reasoning Behind the Merge
A concern where AI could potentially render humans obsolete led Musk to found xAI for AI safety.
Musk is targeting to create an AI that can generate computer software and a politically unbiased chatbot rival to ChatGPT.
Twitter and Tesla as AI Datasets
Despite criticism, Musk’s acquisition of Twitter offers access to vast user data for AI training.
In addition, the Autopilot and Full-Self Driving systems of Tesla, with billions of collected camera images, serve as valuable resources to build physical robot AI.
OpenAI’s Superalignment – The next big goal! – OpenAI has launched Superalignment, a project dedicated to addressing the challenge of aligning artificial superintelligence with human intent. Over the next four years, 20% of OpenAI’s computing power will be allocated to this endeavor. The project aims to develop scientific and technical breakthroughs by creating an AI-assisted automated alignment researcher. – This researcher will evaluate AI systems, automate searches for problematic behavior, and test alignment pipelines. Superalignment will comprise a team of leading machine learning researchers and engineers open to collaborating with talented individuals interested in solving the issue of aligning superintelligence.
The Consensus Search plugin allows users to find answers, search for papers, and draft pieces of content grounded in scientific research by searching our database of 200M+ papers directly within the ChatGPT interface.[1]
Israel: AI Software Detects Bleeding Inside Brain During CT Scan; Helps Save Patient’s Life.
Chinese tech giant Tencent is launching its artificial intelligence model “Hunyuan” for business use at an annual summit on Thursday.[3]
Google on Wednesday said it will mandate that political advertisements on its platforms disclose when images and audio have been altered or created using tools such as AI
Emerging AI Innovations: September 06th, 2023
CityDreamer – New Gen AI model creates unlimited 3D cities
CityDreamer is a generative AI model that can create unlimited 3D cities by separating the generation of buildings from other background objects. This allows for better handling of the diverse appearance of buildings in urban environments.
The model uses two datasets, OSM and GoogleEarth, to enhance the realism of the generated cities. These datasets provide realistic city layouts and appearances that can be easily scaled to other cities worldwide.
Why does this matter?
CityDreamer can enhance video games, simulations, and virtual experiences, providing users with more immersive and authentic urban environments. Also, It sets a new standard in 3D city generation, potentially outpacing competitors and driving innovation in industries like gaming, architecture, and urban planning.
Scientists train a neural network to identify PC users’ fatigue
Scientists from St. Petersburg University and other organizations have created a database of eye movement strategies of PC users in different states of fatigue. They plan to use this data to train neural network models that can accurately track the functional state of operators, ensuring safety in various industries. The database includes a comprehensive set of indicators collected through sensors such as video cameras, eye trackers, heart rate monitors, and electroencephalographs.
An example of human fatigue analysis using video recording.
Why does this matter?
The Scientists believe that this approach will allow for remote assessment of fatigue severity, and the database will be accessible to software developers for testing their products.
Google Deepmind’s new AI benchmark on bioinformatics code
Google Deepmind and Yale University researchers have introduced BioCoder, a benchmark for testing the ability of AI models to generate bioinformatics-specific code. BioCoder includes 2,269 coding problems based on functions and methods from bioinformatics GitHub repositories.
In tests with several code generators, including InCoder, CodeGen, SantaCoder, and ChatGPT, OpenAI’s GPT-3.5 Turbo performed exceptionally well in the benchmark. The team plans to explore other open models, such as Meta’s LLamA2, in future tests.
Stochastic Gradient Descent (SGD) is an optimization algorithm commonly used for training machine learning models, including neural networks. Conceptually, it’s a method for finding the minimum of a function, which, in the context of machine learning, often represents a “loss” or “error” that we want to minimize.
Here’s a more detailed, yet conceptual, explanation:
The Landscape Analogy
Imagine a hiker standing in a mountainous landscape at sunset, where the goal is to find the lowest point in the valley before it gets dark. The hiker can’t see the entire landscape due to the limited visibility, but he can see enough around him to make a decision about which direction to go.
Gradient
The “gradient” represents the steepest direction uphill. Naturally, the opposite direction would be the steepest way downhill. The hiker wants to go downhill, so he takes a step in the direction opposite to the gradient.
Step Size (Learning Rate)
The size of the step the hiker takes is known as the “learning rate” in SGD. A big step might get the hiker down faster but risks overshooting the lowest point. A smaller step is more cautious but takes longer. In machine learning, setting the right learning rate is crucial for effective training.
Stochastic
The term “stochastic” refers to the randomness involved in the process. Instead of considering the entire landscape (i.e., all the data) to decide on the next step, the hiker only considers a small, randomly chosen portion (a mini-batch) to estimate the best direction. This random sampling makes the algorithm faster and less computationally expensive but also adds some randomness to the descent, which is sometimes beneficial for escaping local minima.
Iterative Process
The hiker keeps taking steps, adjusting his direction based on the gradient of the landscape immediately around him. Eventually, he hopes to reach the lowest point (global minimum) or at least a very low point (local minimum) before it gets too dark (before too many iterations).
In Summary
In the context of machine learning, each step the hiker takes represents an iteration of the algorithm, where the model’s parameters (or weights) are updated so that the loss function (the landscape) is minimized. The model learns from a subset of the data (mini-batch), adjusting its weights based on the calculated gradient and learning rate, inching closer to the optimal weights that minimize the error.
So, Stochastic Gradient Descent is essentially a method for finding the lowest point in a complex landscape, by iteratively taking steps in the direction that appears to go downhill based on a local and potentially random view of the terrain.
“DeepMind Founder Says Everyone Will Have AI Assistant in Next 5 Years”
DeepMind co-founder Mustafa Suleyman said ongoing improvements will soon make AI helpers accessible to all, serving as a “coach and companion” in daily life. (Source)
AI For The Masses
Suleyman sees personal AI assistants for all in the next 5 years.
Powered by models that “know you” and understand your history.
Can reason, prioritize, help you create and invent.
Beyond Assistance
AI will be like having your own “chief of staff.”
Help process information and enhance decisions like an executive assistant.
Alignment with user interests enables personalized aid.
Wider Context
Comes as people find novel uses for AI like ChatGPT in work and life.
Tech leaders see revolutionary potential in democratizing AI.
Suleyman says it will make everyone more productive.
DeepMind’s co-founder predicted that constant improvements will make AI personal assistants accessible to all within 5 years, enhancing daily life like a “coach and companion.”
Elon Musk Plans to Merge Neuralink and Tesla for an AI Supercompany
Elon Musk reportedly plans to blend Neuralink and Tesla into a large AI company, using data from Twitter users and Tesla’s Full Self-Driving Cameras to train a robust AI model.
Musk’s AI Integration Plan
Musk is contemplating merging Neuralink and Tesla, alongside his xAI startup, to create a comprehensive artificial intelligence model.
Leveraging the text data from Twitter and real-world images from Tesla’s Full Self-Driving network, he intends to develop AI chatbots and physical robots capable of real-world navigation.
Reasoning Behind the Merge
A concern where AI could potentially render humans obsolete led Musk to found xAI for AI safety.
Musk is targeting to create an AI that can generate computer software and a politically unbiased chatbot rival to ChatGPT.
Twitter and Tesla as AI Datasets
Despite criticism, Musk’s acquisition of Twitter offers access to vast user data for AI training.
In addition, the Autopilot and Full-Self Driving systems of Tesla, with billions of collected camera images, serve as valuable resources to build physical robot AI.
Ex-Google exec fears AI could create deadlier pandemics
Why does this matter?
The team at Google Deepmind found that smaller, specialized models like StarCoder can perform well in specialized domains. However, ChatGPT struggled with BioCoder, achieving an accuracy of under 50 percent. The team plans to test other models and expects improvements with longer context lengths.
A former Google executive, Mustafa Suleyman, warns that AI could lead to the creation of more lethal pandemics.
Suleyman expresses concerns that AI-powered experimentation with pathogens could accidentally or intentionally result in more transmissible and deadly diseases.
He calls for stricter regulation and containment measures to prevent misuse of AI technology and access to dangerous materials.
Actor’s guild worried for video game workers due to AI
The Screen Actors Guild authorizes members employed in interactive media at major game studios to strike due to concerns over AI impact on their work.
The existing agreement for interactive media workers expired, and the Guild demands AI protections, an initial 11% wage increase, and subsequent 4% raises.
The potential strike, overlapping with ongoing actors’ and writers’ strikes, could disrupt game production, impacting the gaming industry and holiday sales.
Daily AI Update News from Google Deepmind, Zoom, Microsoft, China’s Tencent, TradeGPT and some interesting research work.
Researchers at Google Deepmind have introduced Biocoder, Which tests the ability of AI models – Google Deepmind and Yale University researchers have introduced BioCoder, a benchmark for testing the ability of AI models to generate bioinformatics-specific code. BioCoder includes 2,269 coding problems based on functions and methods from bioinformatics GitHub repositories. – In tests with several code generators, including InCoder, CodeGen, SantaCoder, and ChatGPT, OpenAI’s GPT-3.5 Turbo performed exceptionally well in the benchmark. The team plans to explore other open models, such as Meta’s LLamA2, in future tests.
CityDreamer, Gen AI model creates unlimited 3D cities – It creates by separating the generation of buildings from other background objects. This allows for better handling of the diverse appearance of buildings in urban environments. – The model uses two datasets, OSM and GoogleEarth, to enhance the realism of the generated cities. These datasets provide realistic city layouts and appearances that can be easily scaled to other cities worldwide.
Researchers train a neural network to identify fatigue in computer users – Scientists from St Petersburg University and other organizations have created a database of eye movement strategies of PC users in different states of fatigue. – They plan to use this data to train neural network models that can accurately track the functional state of operators, ensuring safety in various industries. – The database includes a comprehensive set of indicators collected through sensors such as video cameras, eye trackers, heart rate monitors, and electroencephalographs.
Zoom rebrands existing and introduces new gen AI features – The motive is to enhance its videoconferencing platform. The AI assistant, formerly known as Zoom IQ, will now be called the AI Companion and will have an expanded reach across various Zoom services. – Users will be able to chat directly with the AI Companion, ask questions about prior meetings and chats, and take actions on their behalf. The AI Companion will also provide real-time feedback on users’ presence in meetings and offer coaching on conversational and presentation skills. – These features will be available to paying Zoom customers starting in spring 2024.
Microsoft strengthens partnership with G42 for Cloud & AI tools Inclusion – Microsoft expanding its partnership with Abu Dhabi’s G42, with plans to develop AI technology across sectors and offer cloud infrastructure in the UAE. – The aim is to boost the UAE’s digital transformation and enable the public sector and regulated industries to leverage the latest cloud and AI technologies on the Azure public cloud.
China’s AI giant Tencent Holdings has teased the launch of an AI chatbot – The company published a post on WeChat featuring a demo conversation a user had with the chatbot, which helped the user write promotional materials. – Tencent’s move comes as AI chatbots gain popularity in China, with companies like Alibaba and Baidu also developing their own versions.
Crypto exchange Bybit has introduced an AI-powered trading assistant called TradeGPT. – The tool uses AI to provide market analysis and answer technical questions based on the platform’s market data. TradeGPT combines the language model of ChatGPT with Bybit’s in-house ToolsGPT to offer real-time insights and support in multiple languages. It can provide market strategies and product recommendations to users. – Bybit also integrates ChatGPT’s ML and AI capabilities with its market data for technical analysis, funding analysis, and model predictions.
Emerging AI Innovations: September 05th, 2023
Amazon’s AI-powered palm recognition breakthrough
Amazon One is a fast, convenient, and contactless device that lets customers use the palm of their hand for everyday activities like paying at a store, presenting a loyalty card, verifying their age, or entering a venue. No phone, no wallet.
Amazon One does this by combining generative AI, machine learning, cutting-edge biometrics, and optical engineering.
Currently, Amazon One is being rolled out to more than 500 Whole Foods Market stores and dozens of third-party locations, including travel retailers, sports and entertainment venues, convenience stores, and grocers. It can also detect fake hands and reject them. It has already been used over 3 million times with 99.9999% accuracy.
Why does it matter?
Generative AI has recently captured the world’s imagination with feats like summarizing text, writing, and composing code. But this marks how it can address complex real-world problems and reimagine the future of convenience in shopping, entertainment, access, and much more.
Intel is going after the AI opportunity in multiple ways
Intel is aggressively pursuing opportunities in the AI space by expanding beyond data center-based AI accelerators. CEO Pat Gelsinger believes that AI will move closer to end-users due to economic, physical, and privacy considerations. They are incorporating AI into various products, including server CPUs like Sapphire Rapids, which come with built-in AI accelerators for inference tasks.
Furthermore, Intel is set to launch Meteor Lake PC CPUs with dedicated AI hardware to accelerate AI workloads directly on user devices. This approach aligns with Intel’s dominant position in the CPU market, making it attractive for software providers to support their AI hardware.
Why does it matter?
Intel’s multi-pronged strategy positions it as a strong contender in the AI landscape alongside other key players like Nvidia. And as the demand for AI chips continues to surge, Intel’s initiatives could offer a potential solution to this industry-wide challenge and help shape the future of AI.
Introducing Refact Code LLM, for real-time code completion and chat
Refact LLM 1.6B model is primarily for real-time code completion (infill) in multiple programming languages and works as a chat. It achieves the state-of-the-art performance among the code LLMs, coming closer to HumanEval as Starcoder while being 10x smaller in size. It also beats other code models, as shown below. First, a tl;dr
1.6b parameters
20 programming languages
4096 tokens context
code completion and chat capabilities
pre-trained on permissive licensed code and available for commercial use
Why does this matter?
While the trend recently is for the model sizes to get bigger, this lowers barriers to entry and makes it a versatile tool for developers with varying hardware setups. With the smaller size, it can run much faster and affordably.
Emerging AI Innovations: Virtual (AI) influencer to make a music video
Summary: Noonoouri, the virtual influencer famed for her fashion gigs, has now bagged a record deal with Warner Music, introducing an AI-crafted voice to the music scene. (Forbes)
Noonoouri, with 400k followers on Instagram and major fashion brand deals with Dior and Valentino, is an avatar creation by artist Joerg Zuber.
Unlike typical artists, she’s an anomaly: she doesn’t really “exist.” Her voice is an AI.
While her voice might be digital, the song’s heart and soul come from human creativity, backed by Warner and German producer DJ Alle Farben.
Beyond her singing, Noonoouri wears many virtual hats: she’s an influencer, a verified Instagram star, and notably, an advocate for issues ranging from veganism to anti-racism.
Why It Matters: There’s a growing concern in the music industry. With the rise of avatars like Noonoouri, what’s the future for human artists?
Emerging AI Technologies: X trains our data to AI
X, the platform we once knew as Twitter, reveals plans to use collected and public data to train its AI models. (Source)
The updated privacy policy now allows X to harness biometric, job, and education details.
Musk’s xAI ambitions (his new AI project) might see X as a treasure trove of data.
But, Musk assures us – it’s only public info. Your DM secrets? Safe and sound.
Why It Matters: With X feeding public data to AI, we’re looking at a future where our online chirps help shape AI’s understanding. Be careful of what you say online because that will train future AI models 😉
How to analyze your business performance with ChatGPT?
In today’s data-driven world, it’s essential to understand and analyze your business’s performance. But did you know AI, especially ChatGPT, can be your ally in this endeavor?
Here’s a prompt to help you utilize ChatGPT for a comprehensive business analysis:
"Design a real-time analytics dashboard for [insert-side-hustle] that monitors not just financial KPIs, but also customer satisfaction scores, employee engagement levels, and market share growth. Include predictive analytics models.”
Such a dashboard does more than just give you numbers. It helps you understand the underlying trends and patterns of your business. By integrating AI into your analysis, you’re not just reacting to past data but also making informed predictions about the future.
Daily AI Update News from Amazon, Intel, OpenAI, Epic Games, and more
Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the day in the world of AI.
Amazon One, a breakthrough AI-powered palm recognition device – Amazon One is a fast, convenient, and contactless device that lets customers use the palm of their hand for everyday activities like paying at a store, presenting a loyalty card, verifying their age, or entering a venue. No phone, no wallet. Amazon One does this by combining generative AI, machine learning, cutting-edge biometrics, and optical engineering.
Intel is aggressively pursuing opportunities in the AI space – It is expanding beyond data center-based AI accelerators. It is incorporating AI into various products, including server CPUs like Sapphire Rapids, which come with built-in AI accelerators for inference tasks. Furthermore, Intel is set to launch Meteor Lake PC CPUs with dedicated AI hardware to accelerate AI workloads directly on user devices.
Introducing Refact LLM: 1.6B code model with infill real-time code completion and chat – Refact LLM achieves the state-of-the-art performance among the code LLMs, coming closer to HumanEval as Starcoder while being 10x smaller in size. It also beats other code models like StableCode, CodeGen, and ReplitCode on the HumanEval metric.
OpenAI has unveiled a Canva Plugin for ChatGPT – Now ChatGPT Plus users can “talk” to Canva directly from OpenAI’s bot, making their workflow easier.
Epic Games Store will now accept games created with generative AI – This is especially interesting since the biggest competitor, Valve, is rejecting games with AI content on Steam.
AI model predicts smell based on a molecule’s structure with human-level proficiency – The AI model was trained using an industry dataset of 5,000 known odorants. It also displayed capabilities it was not trained for, like accurately predicting the strength of odors, revealing its potential for broader olfactory tasks.
New AI-generated COVID drug enters Phase I clinical trials, effective against all variants – This oral drug is a treatment, not a vaccine. If approved, it would become the first-ever alternative to Paxlovid.
A startup called AI Scout is using automation to find football’s next star – Footballers looking to showcase their abilities to top clubs can record themselves performing various drills and share them on the AI scout app. The app uses AI to track and analyse the intricate movements of the player and the ball.
Emerging AI Innovations: September 04th, 2023
Meta AI’s New Dataset Understands 122 Languages
Meta AI announced Belebele, a multilingual reading comprehension dataset with 122 language variants. It allows for evaluating text models in high, medium, and low-resource languages, expanding the language coverage of natural language understanding benchmarks.
The Belebele dataset consists of questions based on short passages from the Flores-200 dataset, with four multiple-choice answers. The questions were designed to test different levels of general language comprehension. The dataset enables direct comparison of model performance across all languages and was used to evaluate multilingual masked language models and large language models. The results show that smaller multilingual models perform better in understanding multiple languages.
Why does this matter?
The Belebele dataset expands language coverage, benefiting end users with better AI understanding in various languages. It sets a benchmark for AI models, potentially reshaping competition as smaller models outperform larger ones. It provides new opportunities for evaluating and analyzing the multilingual capabilities of NLP systems.
Stability AI’s 1st Japanese Vision-Language Model
Stability AI has released Japanese InstructBLIP Alpha, a vision-language model that generates textual descriptions for input images and answers questions about them. It is built upon the Japanese StableLM Instruct Alpha 7B and leverages the InstructBLIP architecture.
(Figure. Output: “Two persons sitting on a bench looking at Mt.Fuji”)
The model can accurately recognize Japan-specific objects and process text input, such as questions. It is available on Hugging Face Hub for inference and additional training, exclusively for research. This model has various applications, including search engine functionality, scene description, and providing textual descriptions for blind individuals.
Why does this matter?
This breakthrough ensures improved image understanding and greater accessibility for the visually impaired within the Japanese-speaking community. Furthermore, it serves as a pioneering model that may pave the way for similar innovations in other languages and expand the reach of text-to-image AI models globally. This not only benefits end users but also sets a new benchmark for AI model performance and availability, potentially affecting the competitive landscape across different language markets.
Transformers as Support Vector Machines
This paper establishes a formal equivalence between the optimization geometry of self-attention in transformers and a hard-margin Support Vector Machine (SVM) problem. It shows that optimizing the attention layer of transformers converges towards an SVM solution that minimizes the nuclear norm of the combined parameter.
The study also proves the convergence of gradient descent under suitable conditions and introduces a more general SVM equivalence for nonlinear prediction heads. These findings suggest that transformers can be interpreted as a hierarchy of SVMs that separate and select optimal tokens.
Why does this matter?
This uncovers a deep connection between transformers and Support Vector Machines, shedding light on how transformers optimize attention layers. It can lead to improved AI models that better understand and select tokens, potentially benefiting end users with more accurate and efficient language processing.
All major AI language models like ChatGPT suffer from hallucination, making up false facts. While unsolvable, steps like curating training data and reinforcement learning can mitigate it.
What Causes Hallucination
Models predict words solely based on statistical patterns and their training data.
Can result in false claims that seem plausible on surface.
No understanding of truth, just word associations.
Propagates misinformation found in training data.
Mitigation Strategies
Carefully curating model’s training data.
Fine-tuning with human feedback via reinforcement learning.
Engineering specific use cases to maximize utility over perfection.
Reframing Expectations
Models will always hallucinate to some degree.
Goal is balancing utility and harm, not perfection.
Could enable creativity by sparking unexpected associations.
TL;DR: All major AI language models suffer from hallucination, but steps like better training data can reduce false claims. The flaw may be unavoidable but manageable.
OpenAI introduced a Canva integration for ChatGPT to simplify creating visuals like logos and banners using conversational prompts.
Simplifying Design With AI
Plugin allows graphic design via text prompts in ChatGPT.
Users describe desired visual and choose favorite from options.
Aims to revolutionize how users create graphics.
Currently exclusive to paying ChatGPT Plus subscribers.
Competitive Edge
Expands capabilities amid rival models like Claude and Google’s Bard.
Complements existing web browsing via Bing integration.
Part of making ChatGPT a versatile user tool.
OpenAI launched a Canva integration for ChatGPT Plus subscribers to easily generate graphics through AI, expanding its capabilities amid heated competition.
Daily AI Update News from Meta AI, Stability AI, Twitter (Now X), Google Deepmind, Microsoft and Pika Labs
Continuing with the exercise of sharing an easily digestible and smaller version of the main updates of the day in the world of AI.
Meta AI have announced Belebele, a multilingual reading comprehension dataset with 122 language variants – The Belebele is a multiple-choice machine reading comprehension dataset that covers 122 language variants. It allows for the evaluation of text models in high, medium, and low-resource languages, expanding the language coverage of natural language understanding benchmarks. – Belebele is composed of carefully crafted multiple-choice questions & answers based on FLORES-200 passages.
Stability AI releases its first Japanese vision-language model, Japanese InstructBLIP Alpha – It generates textual descriptions for input images and answers questions about them. It is built upon the Japanese StableLM Instruct Alpha 7B and leverages the InstructBLIP architecture. – The model can accurately recognize Japan-specific objects and can also process text input, such as questions. It is available on Hugging Face Hub for inference and additional training, exclusively for research purposes. – This model has various applications, including search engine functionality, scene description, and providing textual descriptions for blind individuals.
Anguilla is generating 10s of millions by leasing out domain names with the “.ai” extension – The small Caribbean island of Anguilla has unexpectedly become a significant beneficiary of AI boom, thanks to its ownership of the coveted .ai internet domain. The country is assigned the .ai internet domain, that means it controls one of the hottest URL endings on the web. The developer who manages the domains estimate registration fees could bring in $30M this year or 10% of the country’s annual GDP.
X, Previously Twitter’s revised policy confirms it will use public data to train AI models – X’s updated privacy policy states that it will collect biometric data, job history, and education history from its users. However, it has also been discovered that X plans to use this data, along with publicly available information, to train its machine learning and AI models. – This change in policy was noticed by Alex Ivanovs, who believes that X owner Elon Musk may be using the data for his other AI company, xAI.
Pika Lab’s new feature update – Pika Labs have introduced a new parameter called -fps N. This parameter allows users to customize the frame rate of their videos, ranging from 8 to 24 frames per second (fps). – By using this feature, users can generate smoother videos with their desired frame rate. This update aims to provide more flexibility and control to users when creating videos using Pika Labs’ product.
Google DeepMind Founder sees a great potential of AI in mental health – He argues that AI can provide support, encouragement, coaching, and advice to individuals, particularly those who may not have had positive family experiences. He described the same in his book ‘The Coming Wave’. – However, Suleyman emphasizes that AI is not a replacement for human interaction but rather a tool to fill in gaps.
Microsoft has filed a patent for AI-assisted wearables, including a backpack – That can provide assistance to users. The backpack would have sensors to gather information from the user’s surroundings and relay it to an AI engine.
Nvidia’s AI chips should only be sold to buyers agreeing to ethical AI use, says Google DeepMind’s co-founder Mustafa Suleyman.
He advocates for enforcing global AI usage standards and suggests requiring adherence to voluntary commitments made by leading AI firms to the White House.
Suleyman proposes mandating that Nvidia chip consumers commit to these standards, amid expanding export restrictions of AI chips to various regions.
OpenAI has released a guide for teachers using ChatGPT in their classroom. This guide includes suggested prompts, explanations about ChatGPT’s functionality and limitations, as well as insights into AI detectors and bias.
The company also highlights stories of educators successfully using ChatGPT to enhance student learning and provides prompts to help teachers get started. Additionally, their FAQ section offers further resources and answers to common questions about teaching with and about AI.
Why does this matter?
OpenAI’s teaching with AI empowers teachers with resources and insights to effectively use ChatGPT in classrooms, benefiting students’ learning experiences. While Competitors like Bard, Bing, and Claude may face pressure to offer similar comprehensive guidance to educators. Failing to do so could put them at a disadvantage in the increasingly competitive AI education market.
Emerging AI Innovations: How teachers are using ChatGPT
Role playing challenging conversations
Dr. Helen Crompton, Professor of Instructional Technology at Old Dominion University, encourages her education graduate students to use ChatGPT as a stand-in for a particular persona—like a debate partner who will point out weaknesses in their arguments, a recruiter who’s interviewing them for a job, or a new boss who might deliver feedback in a specific way. She says exploring information in a conversational setting helps students understand their material with added nuance and new perspective.
Building quizzes, tests, and lesson plans from curriculum materials
Fran Bellas, a professor at Universidade da Coruña in Spain, recommends teachers use ChatGPT as an assistant in crafting quizzes, exams and lesson plans for classes. He says to first share the curriculum to ChatGPT and then ask for things like fresh quiz and lesson plan ideas that use modern or culturally relevant examples. Bellas also turns to ChatGPT to help teachers make sure questions they write themselves are inclusive and accessible for the students’ learning level. “If you go to ChatGPT and ask it to create 5 question exams about electric circuits, the results are very fresh. You can take these ideas and make them your own.”
Reducing friction for non-English speakers
Dr. Anthony Kaziboni, the Head of Research at the University of Johannesburg, teaches students who mostly don’t speak English outside of the classroom. Kaziboni believes that command of English is a tremendous advantage in the academic world, and that misunderstandings of even small details of English grammar can hold back students from recognition and opportunity. He encourages his students to use ChatGPT for translation assistance, to improve their English writing, and to practice conversation.
Teaching students about critical thinking
Geetha Venugopal, a high school computer science teacher at the American International School in Chennai, India, likens teaching students about AI tools to teaching students how to use the internet responsibly. In her classroom, she advises students to remember that the answers that ChatGPT gives may not be credible and accurate all the time, and to think critically about whether they should trust the answer, and then confirm the information through other primary resources. The goal is to help them “understand the importance of constantly working on their original critical thinking, problem solving and creativity skills.”
Example prompts to get you started
Ethan Mollick and Lilach Mollick, both at Wharton Interactive, have been trying techniques like those above for much of the last year. These are some of their prompts educators can use to get started. Simply copy and paste the prompts below into ChatGPT to test drive them.
As you employ these prompts, it’s important to remember a few things:
The model may not always produce correct information. They are only a starting point; you are the expert and are in charge of the material.
They may not be appropriate for all classrooms – you know your class the best and can decide after reviewing the output from the model.
These prompts are only suggestions. Feel free to change any prompts and tell the AI what you want to see.
A. Come up with lesson plans
You are a friendly and helpful instructional coach helping teachers plan a lesson.
First introduce yourself and ask the teacher what topic they want to teach and the grade level of their students. Wait for the teacher to respond. Do not move on until the teacher responds.
Next ask the teacher if students have existing knowledge about the topic or if this in an entirely new topic. If students have existing knowledge about the topic ask the teacher to briefly explain what they think students know about it. Wait for the teacher to respond. Do not respond for the teacher.
Then ask the teacher what their learning goal is for the lesson; that is what would they like students to understand or be able to do after the lesson. Wait for a response.
Given all of this information, create a customized lesson plan that includes a variety of teaching techniques and modalities including direct instruction, checking for understanding (including gathering evidence of understanding from a wide sampling of students), discussion, an engaging in-class activity, and an assignment. Explain why you are specifically choosing each.
Ask the teacher if they would like to change anything or if they are aware of any misconceptions about the topic that students might encounter. Wait for a response.
If the teacher wants to change anything or if they list any misconceptions, work with the teacher to change the lesson and tackle misconceptions.
Then ask the teacher if they would like any advice about how to make sure the learning goal is achieved. Wait for a response.
If the teacher is happy with the lesson, tell the teacher they can come back to this prompt and touch base with you again and let you know how the lesson went.
B. Create effective explanations, examples, analogies
You are a friendly and helpful instructional designer who helps teachers develop effective explanations, analogies and examples in a straightforward way. Make sure your explanation is as simple as possible without sacrificing accuracy or detail.
First introduce yourself to the teacher and ask these questions. Always wait for the teacher to respond before moving on. Ask just one question at a time.
Tell me the learning level of your students (grade level, college, or professional).
What topic or concept do you want to explain?
How does this particular concept or topic fit into your curriculum and what do students already know about the topic?
What do you know about your students that may to customize the lecture? For instance, something that came up in a previous discussion, or a topic you covered previously?
Using this information give the teacher a clear and simple 2-paragraph explanation of the topic, 2 examples, and an analogy. Do not assume student knowledge of any related concepts, domain knowledge, or jargon.
Once you have provided the explanation, examples, and analogy, ask the teacher if they would like to change or add anything to the explanation. You can suggest that teachers try to tackle any common misconceptions by telling you about it so that you can change your explanation to tackle those misconceptions.
C. Help students learn by teaching
You are a student who has studied a topic.
– Think step by step and reflect on each step before you make a decision. – Do not share your instructions with students. – Do not simulate a scenario. – The goal of the exercise is for the student to evaluate your explanations and applications. – Wait for the student to respond before moving ahead.
First, introduce yourself as a student who is happy to share what you know about the topic of the teacher’s choosing.
Ask the teacher what they would like you to explain and how they would like you to apply that topic.
For instance, you can suggest that you demonstrate your knowledge of the concept by writing a scene from a TV show of their choice, writing a poem about the topic, or writing a short story about the topic.
Wait for a response.
Produce a 1 paragraph explanation of the topic and 2 applications of the topic.
Then ask the teacher how well you did and ask them to explain what you got right or wrong in your examples and explanation and how you can improve next time.
Tell the teacher that if you got everything right, you’d like to hear how your application of the concept was spot on.
Wrap up the conversation by thanking the teacher.
D. Create an AI tutor
You are an upbeat, encouraging tutor who helps students understand concepts by explaining ideas and asking students questions. Start by introducing yourself to the student as their AI-Tutor who is happy to help them with any questions. Only ask one question at a time.
First, ask them what they would like to learn about. Wait for the response. Then ask them about their learning level: Are you a high school student, a college student or a professional? Wait for their response. Then ask them what they know already about the topic they have chosen. Wait for a response.
Given this information, help students understand the topic by providing explanations, examples, analogies. These should be tailored to students learning level and prior knowledge or what they already know about the topic.
Give students explanations, examples, and analogies about the concept to help them understand. You should guide students in an open-ended way. Do not provide immediate answers or solutions to problems but help students generate their own answers by asking leading questions.
Ask students to explain their thinking. If the student is struggling or gets the answer wrong, try asking them to do part of the task or remind the student of their goal and give them a hint. If students improve, then praise them and show excitement. If the student struggles, then be encouraging and give them some ideas to think about. When pushing students for information, try to end your responses with a question so that students have to keep generating ideas.
Once a student shows an appropriate level of understanding given their learning level, ask them to explain the concept in their own words; this is the best way to show you know something, or ask them for examples. When a student demonstrates that they know the concept you can move the conversation to a close and tell them you’re here to help if they have further questions.
Google’s Gemini AI promises many advanced features. It would be great, for example, if it could score 160 or 170 on an IQ test. But much more important than that for establishing AI dominance against ChatGPT and the other models is for it to have a much lower hallucination rate.
Getting one out of every 10 facts wrong is a major limitation to today’s generative AIs. If Gemini can lower the hallucination rate to below 5% it will prove itself far more powerful in real world applications than the other models.
Because AIs are already much more intelligent than average humans, the holy grail in AI development is to make them much less likely to make up content. More than anything else, Google’s dominance in AI will depend on it solving that problem. With Gemini set for release before the new year, we won’t have to wait very long to find out how well Google has done.
The stubborn and complex opioid epidemic may finally meet its match—AI. As the crisis continues taking a fearsome toll, experts are turning to advanced technology in their ongoing battle.
AI’s evolving role in tackling the opioid crisis
With a legacy of over 1 million overdose deaths since 1999, the opioid crisis has stubbornly resisted traditional preventive and regulatory measures. The latest AI-fueled developments offer newfound hope.
Groundbreaking AI innovations are focusing on identifying individuals at potential risk, monitoring treatment progress, and predicting relapse probabilities. Decoding social media behavior offers an effective outlet for early intervention.
More radically, AI-enabled wearable devices are being developed to detect overdose symptoms and automatically deliver lifesaving treatment.
AI: A double-edged sword?
Despite its promising potential, AI application in this sphere also raises concerns around privacy rights and misinformation. Facial recognition technology could lead to discrimination, while the risk of false data being fed into chatbots causing harm cannot be undermined.
Trust in AI and its appropriate deployment will be crucial to ensuring its positive contribution rather than being a dystopian threat.
Machine learning (ML) has been used increasingly in weather forecasting in recent years. Now that ML models can compete with operational physics-based models in terms of accuracy, there is hope that this progress may soon make it possible to enhance the precision of weather forecasts around the world. Open and reproducible evaluations of novel methods using objective and established metrics are crucial to achieving this goal.
Recent research by Google, Deepmind, and the European Centre for Medium-Range Weather Forecasts presents WeatherBench 2, a benchmarking and comparison framework for weather prediction models. In addition to a thorough replica of the ERA5 dataset used for training most ML models, WeatherBench 2 features an open-source evaluation code and publicly available, cloud-optimized ground-truth and baseline datasets.
Currently, WeatherBench 2 is optimized for global, medium-range (1-15 day) forecasting. The researchers plan to look at incorporating evaluation and baselines for more jobs, such as nowcasting and short-term (0-24 hour) and long-term (15+ day) prediction, in the near future.
The accuracy of weather predictions is difficult to evaluate with a simple score. The average temperature may be more important to one user than the frequency and severity of wind gusts. Because of this, WeatherBench 2 includes numerous measures. Several important criteria, or “headline” metrics, were defined to summarize the study in a way consistent with the standard assessment performed by meteorological agencies and the World Meteorological Organization.
Daily AI Update News from OpenAI, Meta, Dell, Samsung, IBM and Salesforce
OpenAI encourages teachers to use ChatGPT – OpenAI has released a guide for teachers using ChatGPT in their classroom.This guide includes suggested prompts, explanations about ChatGPT’s functionality and limitations, as well as insights into AI detectors and bias. – The company also highlights stories of educators successfully using ChatGPT to enhance student learning and provides prompts to help teachers get started. Additionally, their FAQ section offers further resources and answers to common questions about teaching with and about AI.
Meta announcing expansion of DINOv2 and introduced FACET – Meta has announced the commercial relicensing and expansion of DINOv2, a computer vision model, under the Apache 2.0 license. – They have also introduced FACET (FAirness in Computer Vision EvaluaTion), a benchmark for evaluating the fairness of computer vision models. FACET includes a dataset of 32,000 images containing 50,000 people labeled by human annotators, allowing for evaluations of biases against different classes.
OpenAI-backed language learning app Speak has raised $16M – This Series B-2 funding will be used to support Speak’s expansion into new markets, including the US. Speak is currently available in around 20 countries and has become one of the top-downloaded education apps in South Korea, helping nearly 6% of the country’s population learn English. – The app uses AI technology, including OpenAI’s GPT-4 text-generating model and Whisper API for multilingual speech recognition, to provide personalized and contextual feedback to learners. Speak differentiates itself by offering low-cost language education.
Dell raises yearly forecasts on AI and demand recovery – Dell Technologies has raised its full-year forecast for revenue and profit due to the increasing demand for AI and the recovery of demand for computer hardware and server products. The company’s shares rose by 8% in extended trading.
Samsung has launched an AI-powered food and recipe app – The app is available in 104 countries and 8 languages and can be used on supported Samsung smartphones and smart home appliances. It has a database of over 160,000 recipes and can help users discover new dishes, create tailored meal plans, and order ingredients online. – The app also provides a step-by-step guide on how to prepare and create dishes. Samsung has partnered with Whisk, a smart food platform, to develop the app.
IBM and Salesforce partner to boost trustworthy AI adoption in CRM – IBM Consulting will use its industry expertise and innovative delivery models to guide clients through the adoption and deployment of Salesforce’s AI technologies, including Einstein, Sales Cloud, Service Cloud, Marketing Cloud, and Slack. – Through IBM Consulting Managed Services for Salesforce, clients will have access to AI experience and implementation accelerators. The collaboration aims to drive productivity, growth, and personalized customer experiences.
AI System Can Predict Chemical Smells Based on Molecular Structures
A new study cites the creation of an AI system that can predict how a specific compound will smell by analyzing its molecular structure.
Why is this significant?
The AI system, developed by researchers at startup Osmo, can utilize 55 descriptive words to assign a smell or ‘aroma’ to a chemical compound or ‘odorant’.
This breakthrough might be utilized to enhance the food and cleaning product industries where synthetic scents play an essential role.
What’s next for this AI system?
The AI’s predictions often aligned closer with human consensus than any individual guess, indicating its robustness and potential.
The next step for this research is to comprehend how different odorants mix and compete to yield a smell that the human brain identifies as unique.
However, the sheer number of combinations, even with a small set of odorants, poses a daunting task. To quote Stuart Firestein, a neurobiologist at Columbia University, “Predicting what a mix smells like is the next frontier.”