[AI DAILY NEWS RUNDOWN] AI Beats ER Doctors, Amazon Attacks UPS, and Software 3.0 (May 04 2026)

• The study compared OpenAI’s o1-preview model with two attending physicians across 76 real ER cases and three decision stages of patient care.

• At initial ER triage, the model gave the correct diagnosis 67.1% of the time, compared to 55.3% and 50.0% for the two physicians.

• The two separate physician reviewers tasked with scoring couldn’t tell which diagnoses came from the model and which came from the humans.

• In one case, the AI flagged a rare flesh-eating infection in a transplant patient roughly 12 to 24 hours before the treating doctor caught it.

Why it matters: Millions of people are already using AI daily for health questions, but studies like these are showing the usefulness can also flow the other way to the doctors themselves. If a model generations behind is already beating ER doctors, imagine what the frontier could look like inside the patient care process.

Pentagon announces new AI partners

Image source: Images 2.0 / The Rundown

The Pentagon added 8 AI companies to its classified networks while excluding Anthropic, even as the Washington Post reports the new contracts have the same autonomous-weapons and surveillance limits for which Anthropic was blacklisted.

The details:

• The official agreement list names SpaceX, OpenAI, Google, Nvidia, Reflection, Microsoft, AWS, and Oracle as the companies added to classified networks.

• The Department of War said the new deals will “accelerate the transformation toward establishing the U.S. military as an AI-first fighting force”.

• DoD CTO Emil Michael told CNBC that Anthropic’s supply-chain risk label still stands, but called its Mythos model a “separate national security moment.”

  
  
  

• Anthropic’s exclusion comes days after the White House came out against a broader Mythos rollout over compute concerns impacting its own access.

Why it matters: The White House seemingly wants to have its cake and eat it too — both continuing to shun Anthropic while also wanting priority access to its Mythos model despite the blacklist. There are also some interesting names on that list, namely Reflection, which raised $2B from 1789 Capital, a Donald Trump Jr.-backed fund.

Ask.com shuts down after 25 years

• Ask.com has officially closed its doors after 25 years online, with parent company IAC confirming on May 1, 2026 that it shut down its entire search business, ending one of the web’s earliest recognizable search brands.

• Born in the late 1990s alongside Google Search, the platform started with a question-and-answer format and a butler mascot named Jeeves before rebranding to Ask.com, but it steadily lost ground to Google’s ranking systems.

• In a farewell message, IAC said “a very great search must come to an end” and thanked its engineers, designers, and the millions of users, noting that Ask’s natural-language approach foreshadowed today’s conversational search and AI tools.

AI beats doctors on diagnoses in Harvard study

• OpenAI’s o1 model matched or beat board-certified emergency room physicians on diagnosis, triage and next-step care decisions in a new Science study, based on six experiments using real data from a Massachusetts medical centre.

• The model stood out during early-stage triage, where it handled uncertainty better than doctors by making stronger use of unstructured notes and partial information, though both humans and AI improved as more data came in.

• Researchers, including Harvard’s Arjun Manrai and commentators from Flinders University, warned that AI cannot read visual, body language or auditory cues, and called for prospective clinical trials covering safety, equity and cost-effectiveness before wider use.

Anthropic nears $1.5B Wall Street venture

• Anthropic is close to sealing a roughly $1.5bn joint venture with Blackstone, Hellman & Friedman, Goldman Sachs, and General Atlantic that will push Claude into the portfolio companies owned by those Wall Street firms.

• Anthropic, Blackstone, and Hellman & Friedman are each anchoring the deal at about $300m, Goldman Sachs joins as a founding investor at roughly $150m, and General Atlantic and others cover the rest.

• The venture follows OpenAI’s DeployCo, which drew $4bn from five PE firms last month with a 17.5 per cent annualised return guarantee, while Anthropic’s smaller structure has no publicly reported guaranteed returns.

Amazon declares war on UPS and FedEx

• Amazon is opening its freight network to businesses in retail, healthcare, manufacturing and other industries through a new service called Amazon Supply Chain Services, letting them move, store and deliver goods by ocean, road, rail and air.

• The offering taps Amazon’s fleet of more than 100 cargo planes, over 80,000 trailers and 24,000 intermodal containers, plus warehouses, and includes distribution, fulfillment, parcel shipping, two-to-five-day delivery, warehousing and inventory forecasting.

• The push targets the business-to-business shipping market, a high-margin segment for logistics firms, and mirrors the playbook of Amazon Web Services, which started in 2006 as internal infrastructure before becoming the biggest cloud provider.

Musk texts Brockman seeking OpenAI settlement

• OpenAI’s lawyers have asked a federal judge to let Greg Brockman testify about an April 25 text in which Elon Musk floated a settlement, then warned Brockman and Sam Altman would become “the most hated men in America.”

• The defense says the text shows motive and bias rather than settlement value, citing Federal Rule of Evidence 408 and arguing Musk is using the Oakland lawsuit to attack a competitor after launching xAI.

• Judge Yvonne Gonzalez Rogers, not the nine-person advisory jury, will decide liability and remedies, which could include removing Altman and Brockman from leadership, disgorgement to the charity, and unwinding OpenAI’s for-profit conversion.

GameStop bids $56 billion for eBay

• GameStop has made an unsolicited, non-binding offer to buy eBay for $125 per share in a cash-and-stock deal worth about $55.5 billion, a 20% premium over eBay’s Friday closing price of $104.07.

• GameStop CEO Ryan Cohen, who told the Wall Street Journal he wants to turn eBay into a bigger rival to Amazon, has built a 5% stake and secured up to $20 billion in debt financing from TD Bank.

• The bid raises feasibility questions since GameStop’s market value was $12 billion versus eBay’s $46 billion, but Cohen plans to cut $2 billion in yearly costs and pitch GameStop’s 1,600 stores as fulfillment infrastructure.

Anthropic, OpenAI raise billions for PE push:

The world’s leading AI labs are raising billions ($4 billion for OpenAI, $1.5 billion for Anthropic) to help bring AI into private equity portcos. No, it’s not surprising that companies owned by the cost-conscious want to replace some (expensive) humans with (theoretically less expensive) AI. The two model giants are dueling for enterprise market share, and neither appears set to lose a single deal without a fight.

Cerebras targets $115-$125 share price in upcoming IPO:

AI chip company Cerebras’s IPO is making good progress, with a new SEC filing stating it intends to raise more than $3 billion at a price well over $100 per share. For recent investors, the price is a win. For investors who bought in even earlier, the price is an even bigger victory.

European foundation models?

SAP is on a buying spree. The software giant intends to snap up Dremio to help its customers prepare their data for AI agents, and is also buying Prior Labs, a German AI lab building tabular foundation models, or TFMs. Unlike LLMs, TFMs are aces at handling structured business data, which is right up SAP’s lane. Even better, SAP intends to invest “more than €1 billion over the next four years to scale [Prior] into a globally leading frontier AI lab for [structured] data.”

Neural Computers

by John Coogan

There’s been rumblings about the potential for a Neural Computer for years now since the AI boom began. The basic idea is that the computer would have no software whatsoever, essentially just an LLM that generates whatever you need as you are using the device. Karpathy put it this way: “Imagine a device that takes raw videos or audio into basically what’s a neural net and uses diffusion to render a UI that is unique for that moment.”

It feels like we’re starting to see glimpses of this now. I most recently felt it while trying to understand Ryan Cohen’s proposal for GameStop to takeover eBay for $55.5B. I haven’t tracked either company closely, and I wanted to quickly understand how the two companies size up. In a pre-ChatGPT world, I would have pulled stats from Google or Yahoo Finance, maybe copied them into a spreadsheet if I wanted to see them side-by-side (although Google and Yahoo both offer company comparison views, they are always a bit tricky to navigate). Then if I wanted to share the findings, I could screenshot the sheet, or if I was feeling really fancy, design a slide linked to the data. Now this whole process is a single prompt.

Prompt: “Do a bunch of research on GameStop and eBay’s valuation and key financial metrics, things like growth rate, top line, earnings, revenue, valuation, how the multiples fit together. Build a nicely designed side-by-side comparison of the two companies.” – ChatGPT 5.5 Thinking w/ Images

It’s not a perfect result, I’d probably not use red for all of GameStop’s data because red is usually saved for negative numbers. It’s a good start though and I could easily ask for that change. And obviously you could take this a lot further, delivering an updated image every quarter after earnings drops, or every day if you wanted, or for any other two companies, or for anything else you want. I think the end result is fewer dashboards and more ad hoc analyses delivered on demand to answer the exact question you have at the moment.

Karpathy describes this concept as “Software 3.0” in his Sequoia AI Ascent talk and gave an example of shifting from a vibe-coded app to a fully AI workflow for a project that would annotate menus:

MenuGen is this idea where you come to a restaurant, they give you a menu, and there are usually no pictures. I don’t know what any of these things are — usually 30% or 50% of the things, I have no idea what they are.

So I wanted to take a photo of the restaurant menu and get pictures of what those things might look like in a generic sense. I vibe-coded this app that basically lets you upload a photo, and it does all this stuff. It runs on Vercel, re-renders the menu, gives you all the items, uses OCR for all the different titles, uses an image generator to get pictures of them, and then shows it to you.

Then I saw the Software 3.0 version of this, which blew my mind. It was literally: take your photo, give it to Gemini, and say, “Use Nano Banana to overlay the things onto the menu.”

Nano Banana basically returned an image that was exactly the picture of the menu I took, but it actually put into the pixels the different things in the menu. This blew my mind because actually all of my MenuGen is spurious. It’s working in the old paradigm. That app shouldn’t exist.

The Software 3.0 paradigm is a lot more raw. Your neural network is doing more and more of the work. Your prompt or context is just the image, and the output is an image. There’s no need to have any of the app in between.

I have a few takeaways from this:

First, I think it’s exciting for anyone who’s been hesitant to jump into vibe coding. Frontier models are already able to, in 90% of situations, instantiate exactly whatever’s required to solve an actual problem on the fly under the hood, entirely abstracting away code and tools.

Second, I’m reminded of the 2016 USV blog post “Fat Protocols” which argued that unlike “thin protocols” of the web era like HTTP, FTP, etc. (which accrued minimal value), in crypto, Bitcoin, Ethereum, etc. would be “fat protocols,” do a lot of valuable work, and potentially be more valuable than the application layer that enabled interaction with them. There’s still a bunch of complicated market dynamics around value accrual in the AI value chain, but in terms of doing useful work, the models are certainly getting “fatter” every month to use the USV terminology.

Third, there’s still the question of walled garden jumping. The internet isn’t quite “dead” yet, there’s a lot of good information out there, but many, many platforms are fairly locked down, so writing code, puppeteering a browser running on a Mac Mini, or digging through iMessage locally can still require a different workflow, but that’s more of a legal and business discussion than a technical one. The models will continue to find their way over, under, and through any cracks in the walls of the gardens if users ask politely enough.

There’s still a long way to go here, inference is expensive, everything is slow, and models still make odd mistakes (although less and less these days). I’m still enjoying the image output workflow and it feels like moving up a level of abstraction is more qualitatively binary than a slightly higher score on a particular benchmark.

What Else Happened in AI on May 04th 2026?

OpenAI shipped Codex Pets, animated desktop companions that let you track agent progress without switching back to the app.

OpenAI CEO Sam Altman announced that OpenClaw users can use their ChatGPT subscriptions within the agentic tool, taking a stance against Anthropic’s restrictions.

Maryland signed the U.S.’s first ban on AI-driven grocery pricing, with fines up to $25K for stores caught using personalized shopper data to mark up prices.

SAG-AFTRA secured new AI guardrails in its four-year studio deal, with the guild’s negotiator refusing to sign until Hollywood studios made concessions on AI protections.

A Chinese court ruled that replacing a worker with AI does not legally justify firing them, ordering a tech firm to pay wrongful termination damages.

The Information: XAI Shows How Hard It Is to Use a Lot of GPUs at Once

➞ Lambda response: “The xAI “low utilization” story has people mixing up two different metrics”…

WSJ: GameStop Offers to Buy eBay for $56 Billion

Reuters: Cerebras targets $26.6 billion valuation in US IPO as AI chip demand surges

Cofounder Jack Clark says he thinks there’s a 60% chance of AI reaching recursive self-improvement by the end of 2028

Bloomberg: OpenAI Finalizes $10 Billion Joint Venture With PE Firms to Deploy AI

Roon post on Anthropic as an organization that “worships” Claude goes viral

Mayo Clinic: Mayo Clinic AI helps specialists detect pancreatic cancer up to 3 years before diagnosis in landmark validation study

WSJ: OpenAI Wants to Go Public. First Sarah Friar Needs to Get It to Grow Up.

NYT’s Ezra Klein: Why the A.I. Job Apocalypse (Probably) Won’t Happen

NYT: A.I. Is a National Security Risk. We Aren’t Doing Nearly Enough.

WSJ: Why Almost Everyone Loses—Except a Few Sharks—on Prediction Markets

Patrick Collison: Stripe Atlas hits 100,000 all-time incorporations

Rex Salisbury: Erebor hits $1.1 billion in deposits

Charlie Billelo: 0.1% of the accounts on Polymarket have earned 67% of the profits

Politico Europe: EU accused of wasting €20B on AI computing dreams