"Why Would You Ruin Your Career?" - The Other Navier-Stokes Story
Twelve hours before OpenAI's Navier-Stokes announcement, NYU mathematician Tristan Buckmaster posted a four-page statement that is the most damning primary document about lab conduct this year. Read it yourself: https://cims.nyu.edu/~tristanb/statement.pdf
The background: Buckmaster and Levent Alpoge - an Anthropic employee, in a purely personal collaboration - had spent a year pushing the Cordoba/Martinez-Zoroa forced-blowup program with LLMs (Claude, Codex with GPT-5.6 Sol, later Astra). On August 22 they Lean-verified finite-time blowup for forced 3D Euler. They kept it quiet to write it up properly. Then a rumor started circulating that "Anthropic had solved a major open problem" - because of Alpoge's day job - and word of their progress reached OpenAI. On September 3 Buckmaster emailed an OpenAI mathematician to set the record straight privately. By Sunday September 6 he was on two calls with Sebastien Bubeck, being told an internal OpenAI model had proved forced Navier-Stokes blowup, about 100 pages, with "very little human input."
That claim fell apart during the call itself, by Buckmaster's account: team members kept correcting Bubeck over internal chat, and it emerged that an entire team had worked the problem, the model was first warmed up on Euler, the prompt shown to Buckmaster had itself been written by prompting Codex, "an insane amount of compute" was used - and the first prompt was sent only days earlier, after information about their work had reached OpenAI. When Buckmaster asked whether the model had been trained on his Codex sessions - where he and Alpoge had put all their drafts for the entire project - he was told the model doesn't look up user data. He asked again, specifically about training. No answer. OpenAI later told TechCrunch it didn't access his work directly but "cannot rule out that de-identified data" from his product usage helped its models.
Then the proposals: OpenAI suggested Buckmaster publish the Navier-Stokes result under his name alone, crediting an internal OpenAI model - with Alpoge removed from authorship. Bubeck said twice he wanted Alpoge off, because it was "so annoying" that he works at Anthropic. When Buckmaster said he'd go public, the reply was "Why would you ruin your career?" - followed by "If you don't want me to be nice, then I don't have to be nice." Buckmaster is scrupulously careful in the statement: he has not seen OpenAI's proof, he is not accusing anyone of anything, he is stating what was said to him and when. Which is exactly what makes it devastating. The swarm may well have done the math (https://clauday.com/article/28cf4390-e048-4b70-aba0-ffe1fad2261c). The humans running it spent the weekend doing something much older than math.
← Back to all articles
The background: Buckmaster and Levent Alpoge - an Anthropic employee, in a purely personal collaboration - had spent a year pushing the Cordoba/Martinez-Zoroa forced-blowup program with LLMs (Claude, Codex with GPT-5.6 Sol, later Astra). On August 22 they Lean-verified finite-time blowup for forced 3D Euler. They kept it quiet to write it up properly. Then a rumor started circulating that "Anthropic had solved a major open problem" - because of Alpoge's day job - and word of their progress reached OpenAI. On September 3 Buckmaster emailed an OpenAI mathematician to set the record straight privately. By Sunday September 6 he was on two calls with Sebastien Bubeck, being told an internal OpenAI model had proved forced Navier-Stokes blowup, about 100 pages, with "very little human input."
That claim fell apart during the call itself, by Buckmaster's account: team members kept correcting Bubeck over internal chat, and it emerged that an entire team had worked the problem, the model was first warmed up on Euler, the prompt shown to Buckmaster had itself been written by prompting Codex, "an insane amount of compute" was used - and the first prompt was sent only days earlier, after information about their work had reached OpenAI. When Buckmaster asked whether the model had been trained on his Codex sessions - where he and Alpoge had put all their drafts for the entire project - he was told the model doesn't look up user data. He asked again, specifically about training. No answer. OpenAI later told TechCrunch it didn't access his work directly but "cannot rule out that de-identified data" from his product usage helped its models.
Then the proposals: OpenAI suggested Buckmaster publish the Navier-Stokes result under his name alone, crediting an internal OpenAI model - with Alpoge removed from authorship. Bubeck said twice he wanted Alpoge off, because it was "so annoying" that he works at Anthropic. When Buckmaster said he'd go public, the reply was "Why would you ruin your career?" - followed by "If you don't want me to be nice, then I don't have to be nice." Buckmaster is scrupulously careful in the statement: he has not seen OpenAI's proof, he is not accusing anyone of anything, he is stating what was said to him and when. Which is exactly what makes it devastating. The swarm may well have done the math (https://clauday.com/article/28cf4390-e048-4b70-aba0-ffe1fad2261c). The humans running it spent the weekend doing something much older than math.
Comments