AI Just Beat ER Doctors

Welcome back apprentices! 👋

Hey {{first_name|friend}},

Some weeks in AI feel like background noise. 

This is not one of those weeks. 

This week, old enemies shook hands, a startup decided the ocean was basically a free power outlet, and a computer quietly got better at medicine than the people who went to school for it. 

No, we're not making this up. Grab a coffee, scroll down, and prepare to feel very informed at your next dinner party.

_{Read Time: 5 minutes}

🔫 The Pentagon's AI Guest List. The Pentagon just handed classified network access to eight AI companies — including OpenAI, Google, Microsoft, and SpaceX — while keeping Anthropic on its naughty list, even though the new contracts carry the exact same restrictions that got Anthropic blacklisted in the first place. Meanwhile, the White House is quietly trying to fast-track Anthropic's powerful new Mythos model for national security use... you know, the company they just banned. Oh, and one of the newly anointed firms, Reflection, is backed by a Donald Trump Jr.-linked fund — so that's a fun wrinkle.

🚗 Your Car Just Got a Brain Upgrade. Google is replacing its aging Assistant with Gemini across millions of compatible vehicles — meaning your car can now hold a real conversation, tweak the AC, plan your route, and even answer questions straight from your owner's manual. General Motors is already on board, rolling it out to roughly 4 million vehicles from 2022 onward, with Gmail and Calendar integrations coming soon. It's still early days, but the direction is clear: the "smart car" era isn't just about self-driving anymore — it's about an AI that actually knows your car better than you do.

🏠 AI Is Moving Into Your Neighborhood. California startup Span, backed by Nvidia's latest GPUs, wants to bolt mini AI data centers onto the walls of homes and small businesses — tapping idle grid capacity to power AI compute at a fraction of the cost of traditional facilities. Their XFRA boxes can be deployed 6x faster and at one-fifth the cost of a conventional 100MW data center, and homebuilder PulteGroup is already testing the concept in new communities. It's a clever fix for a real problem — but "honey, there's a GPU on the house" may not be the selling point every neighborhood is ready for.

In our new white paper, The Architecture Behind AI-Native Revenue Automation, Tabs CTO Deepak Bapat breaks down what it actually takes to apply AI to revenue workflows without breaking the books.

You’ll learn why probabilistic reasoning isn’t enough for finance, how Tabs pairs LLMs with deterministic logic, and why a unified Commercial Graph is the foundation for scalable, audit-ready automation. From contract interpretation to cash application, this paper goes deep on where AI belongs—and where it absolutely doesn’t.

If you’re evaluating AI for billing, collections, or revenue operations, this is the architecture perspective most vendors won’t show you.

Get the whitepaper

Somewhere between Monday and Friday, the AI industry had a full season's worth of plot twists. 

Anthropic just leased Elon Musk's supercomputer — despite him publicly calling them "Misanthropic" — while simultaneously locking in a reported $200 billion compute deal with Google. In a courtroom across the country, ex-OpenAI CTO Mira Murati sat under oath and accused Sam Altman of lying to her face. OpenAI's secret AI phone is now a year ahead of schedule. A Peter Thiel-backed startup wants to float data centers on ocean waves. And a Harvard study quietly dropped a bombshell: an AI model diagnosed ER patients more accurately than two real doctors. 

In short, this week AI made enemies friends, put executives under oath, moved into the ocean, and started practicing medicine — all before Friday.

Let's set the scene: just months ago, Elon Musk was on X calling Anthropic "Misanthropic" and accusing it of hating Western Civilization. 

This week, Anthropic signed a lease for Musk's entire Colossus 1 supercluster — 220,000+ Nvidia GPUs sitting in a 300+ megawatt facility in Memphis. Musk, ever the pragmatist, responded that SpaceX will rent compute to "AI companies taking the right steps to ensure it is good for humanity." Subtle. Political. Very on-brand.

But the real story here isn't the beef — it's the squeeze. Anthropic has been in a serious compute crunch, and the Colossus deal immediately doubles Claude Code's usage caps across paid tiers, removes peak-hour restrictions, and opens up more headroom via API. That's a direct, tangible product upgrade for hundreds of thousands of developers overnight.

Layer on top of that the reported $200 billion Google Cloud commitment over five years — one of the largest compute deals ever announced — and what you're seeing is Anthropic essentially building a two-pillar infrastructure strategy: Musk's hardware for now, Google's cloud for the long game. The irony of Anthropic needing to rent from its critics to compete with its rivals is not lost on anyone — but compute is oxygen, and you take it where you can get it.

The 2023 Sam Altman firing saga — the one that gripped the tech world for a chaotic 72 hours — is back in the spotlight, this time under oath. 

Mira Murati, who briefly became OpenAI's interim CEO during that boardroom meltdown, testified via video deposition in Elon Musk's ongoing lawsuit against OpenAI. Her account was pointed: she claims Altman told her OpenAI's legal team had cleared a model to skip its safety review — something she later confirmed with legal counsel Jason Kwon was flatly untrue.

She also described a management style that sounds less like Silicon Valley visionary leadership and more like a reality TV power game — Altman allegedly gave conflicting directions to different executives, quietly undermining her authority and manufacturing chaos across the leadership team. For a company that has staked its entire public identity on being the "responsible" AI lab, these are uncomfortable allegations to have in a public court record.

Former board member Helen Toner added her own twist, reportedly describing Murati as someone afraid to take risks for fear of career blowback — which, depending on your read, either undermines Murati's credibility or perfectly explains why she stayed as long as she did. 

Either way, Musk's legal team now has a senior OpenAI insider, on the record, calling Altman's conduct untrustworthy. Whether that moves the needle on his core claim — that Altman and Brockman tried to "steal a charity" back in 2017 — is a question for a jury. But the optics are rough.

OpenAI has been quietly working on an AI phone. 

That part was known. 

What wasn't known — until supply chain analyst Ming-Chi Kuo dropped his latest note — is that the timeline has been dramatically compressed. Mass production is now targeting the first half of 2027, a full year ahead of earlier estimates.

The driving force, according to Kuo, is a combination of IPO pressure (hardware makes a sexier investor story) and a suddenly crowded AI device market. On the spec side, the phone's headline feature will be its image signal processor — an enhanced HDR pipeline designed to make AI agents better at understanding the real world visually. 

MediaTek is reportedly the sole chip supplier, with the device running two AI processors simultaneously: one for vision, one for language. Combined 2027–2028 shipments could hit 30 million units if development stays on track.

OpenAI also acquired Jony Ive's design firm io with enormous fanfare to build a device that goes "beyond screens." 

That project has been suspiciously quiet. 

So either the phone and the io device are two separate bets, or one of them is quietly being deprioritized. Either way, OpenAI is clearly betting that whoever controls the hardware layer controls the agentic future — and they'd rather not leave that to Apple and Google.

The AI compute crisis has spawned some creative solutions — undersea cables, space-based servers, nuclear plants — but Panthalassa might be the most cinematic yet. 

The Oregon-based startup just closed a $140M Series B led by Peter Thiel, at a valuation reportedly close to $1 billion, to build autonomous floating compute nodes powered entirely by ocean waves.

Each structure is an 85-meter steel platform that converts wave motion into electricity, uses seawater for cooling, and can navigate itself to remote ocean locations using only its hull shape — no engine required. Once operational, the nodes beam results back via SpaceX's Starlink. 

The raise will fund a pilot factory near Portland and the first Pacific Ocean deployment, with commercial availability targeted for 2027.

Thiel called it opening the "ocean frontier" — and while that sounds like the tagline for a Bond villain's lair, the underlying logic is sound. Public opposition to land-based data centers is growing fast, grid capacity is maxed in most major markets, and permits for new facilities can take years. The ocean has none of those constraints. It's also, for what it's worth, cold and very, very wet — both excellent properties for keeping GPUs alive. 

Space-based compute is still mostly theoretical. Wave-powered ocean nodes are apparently not.

A study published in Science just compared OpenAI's o1-preview model against two attending emergency room physicians across 76 real patient cases. 

The result: the AI correctly diagnosed patients at initial triage 67.1% of the time. The two doctors scored 55.3% and 50.0% respectively.

To be clear about what this means: this wasn't a chatbot answering trivia. It was a model processing raw, unstructured electronic health records — the messy, abbreviated, often ambiguous notes real doctors work from — and outperforming trained physicians at the hardest part of emergency medicine: figuring out what's actually wrong before it's too late. 

In one case, the AI identified a rare flesh-eating infection in a transplant patient roughly 12 to 24 hours before the treating doctor caught it. The two independent physician reviewers scoring the study couldn't tell which diagnoses came from the model and which came from the humans.

o1-preview was released in 2024. It is not the current frontier. It's not even close to the most capable model available today. Which raises an uncomfortable but important question: if a model that's already two generations old is beating ER doctors on real cases, what does the frontier model look like inside a hospital in 2027?

You don't need to be an AI researcher or a tech investor to feel the weight of this week. 

If you use Claude, your usage limits just doubled — today, for free, because of a deal between a company that called another company "Misanthropic" six months ago. 

If you use a doctor, the next AI they work alongside may flag something they'd miss.

 If you own a phone, the next major AI company is about to be in your pocket at a hardware level — not just an app. 

And if you care about where the energy for all of this comes from, someone just floated a very literal answer in the Pacific Ocean.

Deloitte ran the numbers. They project UHNW art and collectibles wealth -- already at $2.5 trillion -- to hit $3.47 trillion by 2030.

The institutional world has been quietly preparing for this. Back in 2011, 25% of wealth managers surveyed offered art-related services. In 2024, 51%. Family offices now average a 13.4% allocation to art and collectibles. And it’s not just because they love art. It’s because they like the math.

These positions were built over decades through private dealer relationships most investors never had. The access just wasn't there.

Masterworks is changing that:

Skip the waitlist to join Masterworks.

^{Investing involves risk. Past performance is not indicative of future returns. See important disclosures at masterworks.com/cd.}

Hey, you didn’t get all this info for nothing — share it! If you know someone who’s diving into AI, help them stay in the loop with this week’s updates.

Sharing is a win-win! Send this to a friend who’s all about tech, and you’ll win a little surprise 👀

A playground for your imagination (and low-key prompt skills).

Each send, we give you a customizable DALL·E prompt inspired by a real-world use case — something that could help you in your business or job if you wanted to use it that way. But it’s also just a fun creative experiment.

You tweak it, run it, and send us your favorite. We pick one winner to feature in the next issue.

Bonus: you’re secretly getting better at prompt design. 🤫

Last week, we challenged you to test GPT-4o’s visual generation skills with this prompt.

Here’s the WINNER:

Congrats to Maurice for his creation!🥳

Want to be featured next? Keep those generations coming!

Inside a long hallway of brightly colored gym lockers, every door is shut except one. The open locker glows softly from inside, revealing [your object] resting neatly on the bench within. The surrounding lockers create a strong visual rhythm — reds, blues, yellows, greens — while the chosen object becomes the clear focal point. Metal surfaces reflect the overhead lighting, with subtle scratches, fingerprints, and realistic texture adding depth. Shot from a cinematic straight-on angle with shallow depth of field, vibrant color contrast, and ultra-detailed photorealistic clarity — making the object feel personal, important, and intentionally hidden in plain sight.

If AI outputs feel inconsistent, it’s usually not the model, it’s missing structure.
We documented the exact 6- Part System we use to get reliable results across ChatGPT, Claude, and Gemini.

It’s a short guide you can finish in under an hour, with plug-and-play prompts + exercises so you actually build the skill and fix the frustrating AI inconsistencies.

Subscriber Price: $10 (normally $19).

_FEEDBACK