"I’m sorry that I have no words of wisdom or inspiration. I get sad and scared too. I think maybe it’s part of the natural price of wanting to do this kind of work."
~David Foster Wallace
https://t.co/T2gIfa5TWh
What is left out here:
1. Do the doctors actually look?
2. Do patients most need advanced knowledge or just REMEMBERING simple stuff ("Do blood cultures BEFORE antibiotics")
3. Who controls the algorhym?
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at https://t.co/YOtzND0Jqk.
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
One person, 11 labs, an acquisition. The headline is drug discovery, but the story is that one founder ran a company that used to need a whole team.
I run software for several shops the same way. Smaller stakes, same shape.
Is "one-person company" the new default, or still the exception?
Chef opens a restaurant. People really like it. He books out reservations for 3 months.
But now people are upset there’s a 3 month wait to go to the restaurant.
Then, the government decides to put price caps at how much he can charge customers. He still needs to pay overhead and staff and ingredients, though.
So he decides he needs to serve more customers to cover his costs. So he shortens the allowable time for meals. Of course, now the customers aren’t happy about having to rush through their meals. Many of them insist on eating at a leisurely pace, or they send their food back because they didn’t like it, or they show up late for their reservation and insist on being seated anyway.
Meanwhile, the chef is trying to keep up. He feels bad he can’t put the same attention into each meal he did before. But if he doesn’t serve enough people in a day, he will have to close down (or sell to the large restaurant conglomerate down the street).
Oh, and the government also decides that he now needs to do two hours of computer work for every hour he spends cooking. They force him to use a terrible, but expensive, computer system, too, not the one he wanted. So he needs to serve even more customers to cover those costs as well.
The customers are rightly upset. They’re waiting longer and getting less personalized service. The chef is also upset that he’s spending more on overhead but delivering worse meals.
Nobody is happy here.
We've been getting a lot of questions lately about how the Internet Archive digitizes books.
The short answer: page by page, by hand.
You may remember our viral 2021 video of Eliza Zhang scanning a book. That's still how we do it.
Meet Eliza, and learn how we scan books: https://t.co/gt9s5lbPE4
I entered UCLA as a freshman at 46 and graduated with a computer science and digital media management degree at 50. In the ten years since, I’ve made three feature films, two film shorts, established a film festival, written for numerous publications, and published two books (third one on the way). You’re either alive or dead. Those are the only two ages.
Family doctors build a caring relationship with you and your family. They listen, document your health history and help you make informed decisions about your health.
Learn more about how to find the right family doctor: https://t.co/Z8hL4kMmHO
EHRs are a tax on physician time. Countless hours of our lives are lost waiting on screens to load, or the clicks to process.
Many hospitals host remotely, making work happen through a remote desktop, so every single action has a quarter second delay.
It all adds up.
Our perspective on decision authority in health AI was just published at @Nature Health, w/ Aiden Gu
free access https://t.co/JE0OVWFe2l
https://t.co/OBhpEVTPZg
Happy Independence Day!
There’s a story from the end of the Revolutionary War I want to tell as we celebrate America’s 250th Birthday, and it’s one everyone in the world can learn from.
George Washington, at that moment, after commanding the American forces to victory, was the most powerful man in the new country. Many people talked about making him King of America.
Across the ocean, King George was sitting with an American painter, and asked what he thought Washington would do now that the war was ending. The painter said he believed he would go back to his farm.
The King said, “If he does that, he will be the greatest man in the world.”
As the war officially ended, Washington came to speak to Congress and said, “Having now finished the work assigned me, I retire from the great theater of Action.” He returned his commission they’d given him in 1775 - after more than 8 years of leading the Americans to victory without pay, and he was home at Mount Vernon for Christmas.
Of course, he was elected as our first President a few years later, and after two terms, showed the same selflessness again when he willingly gave up his power and went back to Mount Vernon again.
That’s true greatness. He had all the power in the world. But power, alone, does not make you great.
Washington’s greatness came from being a true servant - to a cause much bigger than himself. His greatness was his complete lack of selfishness.
The whole story of American Independence is a story of selflessness. It’s a story of people who set their self-interest aside and worked for each other.
We’ve all heard the line about “We must all hang together, or, most assuredly, we shall all hang separately.”
Apparently, Ben Franklin might have actually never said that.
But that’s fine, because the same mentality is right there in the last line of the Declaration of Independence, published on this day 250 years ago:
“And for the support of this Declaration, with a firm reliance on the protection of divine Providence, we mutually pledge to each other our Lives, our Fortunes and our sacred Honor.”
We mutually pledge to each other.
No one was in this alone. No one was in it for themselves. This was a group of people with different backgrounds who were in it for each other.
Today is a reminder: greatness comes from what we do for each other, never what we do for ourselves.
That’s a lesson that applies no matter what country you call home.
It’s a lesson that doesn’t require any law passed by a politician, because, let’s be honest, if you’re waiting for selfless politicians, I really hope you are not holding your breath.
All of us have the power to be there for the people around us. For our families and friends. For our neighbors. For everyone.
All of us can reach for greatness.
It’s as simple as looking beyond yourself, seeing past the mirror, picking your eyes up from your phone, and pledging to be there for each other.
Happy Fourth. May you all find your own version of greatness today by lifting each other up.
Lift up your neighborhood. Lift up America. Lift up the World.
Dr. Mike Moore (@MikeMooreDO) shares how family physicians are uniquely positioned to provide medical context and advice because of their continuity of care, supporting you through every life stage. 🫶
When I applied to medical school in 1996, lots of people (attendings, residents, some family members) told me the profession was doomed because of HMOs, longer hours, and lower pay. It's been 30 years, and medicine is still in demand, with excellent job security. Don't let the current fatalists get you down. The world will always need doctors.