OpenAI Navier-Stokes claim hit with data allegations — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI announced it solved the Navier-Stokes problem, unsolved for roughly 90 years, using an internal AI model more powerful than GPT-6 Astra alongside 10,000 concurrent agents; the company began training the model on August 28th.
- The Navier-Stokes problem is one of seven Millennium Prize Problems, each carrying a $1 million reward, but OpenAI says it does not plan to claim the prize.
- Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic) published findings on a related problem one day before OpenAI's announcement, after working on their proof using OpenAI's Codex and Anthropic's Claude.
- Buckmaster asked OpenAI whether its model had been trained on his Codex drafts and said he got no answer on the training question; he later wrote on Mastodon that OpenAI is "openly admitting they used training data from a period after we found our result."
- OpenAI stated "no specific user data was accessed" to solve the problem, while staff member Sebastien Bubeck claimed the company did not see the researchers' work until it was publicly released and said the proofs differ significantly; the company added it "cannot rule out that de-identified data derived from their usage of our products helped improve our models."
Why it matters: The timing cuts against OpenAI: Buckmaster and Alpöge published a related result one day before, using OpenAI's own Codex product for their drafts. OpenAI denies accessing specific user data but concedes it "cannot rule out" that de-identified data improved its models — exactly what Buckmaster alleges on Mastodon. The dispute puts the credibility of AI-generated mathematical proofs under a microscope.
Ask SkimNews
