x.com/eigenstate/status/2106038856378352046
1 correction found
according to Tristan Buckmaster we know officially now that internal anthropic models and lots of compute were heavily used in the Navier Stokes collaboration.
Buckmaster’s own statement does not say this. He described the work with Levent Alpöge as a personal collaboration with no official employer involvement, and his “insane amount of compute” remark referred to OpenAI’s separate Navier–Stokes effort, not to their collaboration.
Full reasoning
Buckmaster’s published statement says his work with Levent Alpöge was "a purely personal collaboration, free of any institutional agreements or official involvement by either of our employers". In the same document, he says they "used several LLMs throughout: Anthropic’s Claude, OpenAI’s Codex" and that he paid for the tools from his own research funds. That is not an "official" Buckmaster confirmation that Anthropic internally supported the project or that Anthropic internal models and major Anthropic compute were heavily used.
Buckmaster also does not use his "lots of compute" remark about his own collaboration. The line about "an insane amount of compute" appears in the section describing what OpenAI told him about OpenAI’s separate Navier–Stokes effort. So the post mixes Buckmaster’s description of OpenAI’s project with his separate description of his own collaboration.
OpenAI’s later writeup does contain OpenAI’s characterization that Buckmaster and Alpöge had used "an internal Anthropic model", but that is OpenAI’s statement in the "Concurrent work" section—not Buckmaster’s own published statement. OpenAI’s page also says that what they had produced was a forced Euler result, not a released Navier–Stokes solution by Buckmaster and Alpöge.
So the post is inaccurate in two ways: it attributes the "internal Anthropic model" point to Buckmaster when his own statement does not say that, and it treats Buckmaster’s "insane amount of compute" remark about OpenAI as if it described Buckmaster and Alpöge’s collaboration.
2 sources
- Tristan Buckmaster statement
My work with Levent has been a purely personal collaboration, free of any institutional agreements or official involvement by either of our employers. We used several LLMs throughout: Anthropic’s Claude, OpenAI’s Codex... I pay for the tools my group uses out of my own research funds... it emerged that an entire team had been working on the problem ... and that an insane amount of compute had been used.
- On the Navier–Stokes Millennium Prize Problem | OpenAI
At that point we found out that, using an internal Anthropic model, they had produced a resolution of the forced Euler problem... Across all attempted problems, the agents sent 4.9 million messages... In the process of resolving the Navier–Stokes problem... 130 billion output tokens.