OpenAI claims that one of its artificial intelligence models not yet released has solved the Navier-Stokes problem, one of the seven Millennium Problems indicated by the Clay Mathematics Institute, in just a few dozen hours. The company talks about a milestone for the research, but the discovery has not yet been verified and a dispute has already broken out over the authorship of the result with a team linked to Anthropic and New York University
OpenAI announced that one of its artificial intelligence models, not yet released to the public, would have solved the Navier-Stokes problem in a few dozen hours, one of the seven Millennium Problems: the most difficult open mathematical questions, identified by the prestigious Clay Mathematics Institute and left unanswered for generations. The US company speaks of a turning point, but the result has not yet been confirmed by the scientific community and doubts have already emerged regarding the real authorship of the discovery.
What is the Navier-Stokes problem
The question concerns the equations that describe the movement of fluids, from air to water. They are tools used in aircraft design, weather forecasting and the study of blood flow. According to OpenAI's report, the work of its agents demonstrated that these equations can collapse over time, thus answering the question in the form set by the Millennium Prize. Each of the seven problems is associated with a million dollar reward, but the company has made it known that it will not claim the money if the solution is validated.
“Milestone for AI Research”
“This is a significant milestone for AI research, and its promise for the world is that even more of our most difficult questions can be answered,” Mark Chen, chief research officer at OpenAI, told reporters. Again according to the company's reconstruction, the agents would have reached the solution on Saturday, approximately 88 hours after the start of the project. Chen specified that the calculation costs are around "millions of dollars", a figure equal to about a thousand times that spent on previous mathematical results.
Read also
OpenAi launches GPT-6 Astra: "It's a generational leap"
Ten thousand agents working together
In a blog post describing the discovery, OpenAI explained that it began training a new model in late August that is more capable than any product made public so far. A few days later, faced with online rumors that a competitor had solved some of the Millennium Problems, the company staked the system on all six of the remaining open ones. Work on Navier-Stokes showed unexpected promising results, and enormous computing power was poured into it. In the final phase, researcher Sebastien Bubeck said, ten thousand agents were working on the problem simultaneously, exchanging messages.
The paternity dispute
However, the announcement came a few hours after that of Tristan Buckmaster, a mathematician at New York University, and Levent Alpoge, who works for rival Anthropic. The two had disseminated their work, aided by AI, on three related equations. In a statement published alongside the articles, Buckmaster argued that OpenAI would address the problem only after news of his research had spread, then following the same unusual approach that he and Alpoge had developed over months of work.
"We didn't use their prompts"
OpenAI researchers denied having access to the rival team's work and reportedly offered to share credit, before realizing that Buckmaster and Alpoge had failed to fully solve the problem. “To be clear, we did not use their prompts or demonstrations to target our models or our agents,” said Bubeck, who also rejected what he called false and defamatory allegations against him.
