OpenAI Claims It Solved Navier-Stokes. 25 Fields Medalists Say Stop.

0:00 "Why would you ruin your career?" A mathematician in New York says an Open AI researcher said [music] this to him. Why? Because he asked them to give credit to his research partner. Four days earlier, Open AI announced something huge. 10,000 AI agents had solved one of the hardest math problem in the world in 88 hours. The prize for that problem is $1 million. Since then, 25 of the best mathematicians alive have signed a letter [music] against Open AI. These are Fields Medal winners, biggest prize in mathematics. Open AI also canceled its sponsorship of a university competition, and nobody has checked the proof still. I'm not a mathematician. I work in finance, and that is exactly why this story matters to me.

0:50 Because this is a story about a huge claim that nobody can check. Stay with me. The part that should really worry you is not the math. On the 8th of September, Open AI said it solved the Navier-Stokes problem. Not one model, a group of AI agents, around 10,000 of them. They worked for 88 hours. 73 million people saw that post. So, what is Navier-Stokes? This equation explain how water, air, and blood move. We use them every day for weather, airplane, medicine. And here is the strange part. We have never proved that they always work. It is one of the seven famous unsolved problems.

1:36 Each one is worth $1 million. Only one has ever been solved. Quick thing. Every headline calls this a 90-year-old problem. I tried to find where that number comes from. I couldn't. The equations are 200 years old. The prize is 26. If you know where 90 comes from, tell me in the comments, please. And remember that feeling number everybody repeats that nobody can check. Now, what happens behind the announcement? Be careful here. This is one side of an argument. These are claims from one man, not proven facts. His name is Tristan Buckmaster. He teaches at New York University.

2:22 Since August, he was working on this problem with a partner. His partner is called Lavent Alpok. And this is important. Alpok works at Anthropic, OpenAI's biggest rival. They were close. They were moving fast. Then, OpenAI announced they had solved it. Buckmaster says five things. One, he says OpenAI only started this after they learned he was close. Two, he says their model used the same method he created. Three, he says he asked them a direct question, "Did your model read my notes?" Four, he says an OpenAI researcher gave him a choice, "Remove your partner's name or give up the prize." Five, he says when he refused, that researcher said this, "Why would you ruin your career?" And if you don't want me to be nice, then I don't have to be nice.

3:19 Stop for a second. Think about what that means. Imagine you work on something for years and then someone tells you to remove your partner's name or lose everything. Days later, 25 Fields medal winners signed an open letter. And listen carefully to what they say. They do not say the proof is wrong. Nobody says that. They say it was announced before anyone checked it. They say there is no proper write up. They say other people did not get credit. Then Terrence Tao spoke. Many people call him the greatest living mathematician. He said AI companies use famous problems as his words, marketing proof points.

4:07 In simple English, he thinks they sell a product not to help science. If you want more of this story behind the headline from someone who actually works in finance, subscribe. And if this made you think, leave a like and a comment. Okay, back to it. Quick scoreboard, because this is being reported badly. OpenAI announced the solution. True. Someone has checked it. No. The proof is wrong. Nobody has said that. Here is uncomfortable ending. OpenAI might be completely right. This might be a real historic result, and it could still have been handled badly. 25 of the best mathematicians alive think it was.

4:55 Both can be true, but I keep coming back to what Tao said, marketing proof point, because that is the problem with a result nobody can check. It is not true and it is not false. It is just useful.