OpenAI's announcement that it had solved a nearly century-old math problem touched off a dispute over credit, deepening controversy just a day after the claim.
According to Axios and Fortune, OpenAI said Tuesday (local time) that it had solved the Navier-Stokes equations, one of the seven Millennium Prize Problems posed by the Clay Mathematics Institute. The equations describe the motion of fluids such as water and air and are used in weather forecasting and aircraft design.
The work involved an internal AI model more advanced than GPT-6 Astra, one of OpenAI's AI models. Beginning Sept. 1, about 10,000 AI agents worked simultaneously and reached a conclusion after 88 hours. The agents exchanged 4.9 million messages between them and used about 300 billion output tokens. OpenAI said the effort cost several million dollars.
OpenAI said on its blog that it decided to take on the equation because of rumors that two other Millennium Problems, previously considered unsolved, had already been cracked.
The mathematicians rumored to have solved those Millennium Problems were Tristan Buckmaster, a professor at New York University, and Levent Alpoge, a mathematician at Anthropic. In practice, neither had actually solved a Millennium Problem — they were preparing proofs for equations adjacent to Navier-Stokes.
OpenAI said that after completing its own verification of the Navier-Stokes solution on Sunday, it contacted the two mathematicians and proposed announcing the results together.
Buckmaster said that when contacted, he asked "whether the model had accessed or trained on our Codex (an OpenAI product) sessions." He added, "I was told the model doesn't view user data, but when I asked again whether it had trained on it, I got no answer." The exchange raised the possibility that OpenAI's AI model may have drawn on unpublished research during training.
In response, OpenAI researcher Sebastien Bubeck denied at a press briefing that the company had used the two mathematicians' work or accessed materials shared on their servers. However, OpenAI said it could not rule out the possibility that data from the two mathematicians' use of its products "helped improve our models."
Buckmaster said Bubeck presented two options for handling the announcement. OpenAI could state, in announcing it had solved the problem, that the two mathematicians were entitled to the prize money. Or it could drop Alpoge's name as a co-author, since OpenAI did not want him included given his affiliation with Anthropic. Bubeck rebuffed the account as "false and inflammatory."
Meanwhile, OpenAI has only announced that it solved the problem without releasing the underlying details, while Buckmaster and Alpoge have already published their paper.
A mathematician who solves a Millennium Problem receives $1 million from the Clay Mathematics Institute. OpenAI, which is preparing for a listing at a valuation of $1 trillion, said it has no intention of claiming the prize.
dbsdn1110@heraldcorp.com
