Navier-Stokes – Tristan Buckmaster [pdf]
Points and comments are a snapshot, not live.
A mathematician alleges OpenAI raced to claim a Navier-Stokes result after learning of his work.
Tristan Buckmaster states that he and Levent Alpöge, using LLMs, achieved finite-time blowup results for incompressible porous media, Boussinesq, and 3d incompressible Euler, and believe they have blowup for hypo-dissipative Navier-Stokes. Buckmaster says OpenAI told him an internal model produced a forced Navier-Stokes blowup proof using the same approach, after rumors of their work reached the company. He describes calls where OpenAI proposed publication terms excluding Alpöge due to his Anthropic affiliation, and recounts being told "Why would you ruin your career?" when he declined.
Buckmaster notes their Euler paper is "AI slop" due to pressure to release early. He credits Diego Córdoba and Luis Martínez-Zoroa for the original approach and states Martínez-Zoroa deserves a Fields Medal.
What commenters are saying
The thread centers on Buckmaster's accusation that OpenAI exploited knowledge of his team's partial Navier-Stokes result to produce a competing proof. Several commenters note the dramatic timeline: news of Anthropic's supposed breakthrough spread, then OpenAI quickly claimed a result via the same method. Some question why Buckmaster used OpenAI's models (Codex) for the work if concerned about data use. Others point out the approach involves forced versions, not the full Clay problem, and note that Buckmaster does not claim to know for sure that OpenAI used his data, only that he was told the prompt was sent after information about his work reached them. A quote from Sholto Douglas of Anthropic is cited, calling the data-theft scenario unlikely.