OpenAI announced a solution to the Navier-Stokes problem, a mathematical challenge unsolved for approximately 90 years. The company utilized an internal AI model, described as more powerful than GPT-6 Astra, operating with 10,000 concurrent agents. Training for this model commenced on August 28.
The Navier-Stokes equation is one of the seven Millennium Prize Problems, carrying a $1 million reward. OpenAI stated it does not intend to claim this prize. The company claims its model demonstrated unprecedented performance in mathematical benchmarks.
The announcement sparked controversy involving New York University professor Tristan Buckmaster and Anthropic researcher Levent Alpöge. Buckmaster alleged that OpenAI might have accessed their private research data stored in Codex sessions. OpenAI denied accessing specific user data, though it acknowledged that de-identified data from product usage could have influenced model improvements.
Sebastien Bubeck of OpenAI stated that the company did not review the work of Buckmaster and Alpöge before its public release. He noted that the proofs developed by OpenAI differ significantly from those of the researchers. Buckmaster challenged this, pointing to OpenAI's admission regarding the use of training data from the period following their own discovery.
