OpenAI creates advisory group on AI and mathematics
OpenAI says a new internal model has solved more than 100 open mathematics problems, including the Navier-Stokes problem. To review the impact of these advances, it is creating an independent advisory group made up of mathematicians from institutions such as the IAS, Cambridge, Stanford and Harvard.
OpenAI says a new internal model has solved more than 100 open mathematics problems, including the Navier-Stokes Millennium Prize Problem. The company started training it on August 28 and says the pace of progress surprised even its own mathematicians.
An open problem is a mathematical question with no accepted solution yet. In this case, OpenAI has not publicly identified the model or detailed the problems it solved, so the announcement still depends on independent review.
Why it matters
The Navier-Stokes problem aims to explain precisely how fluids such as water and air move. It is one of the seven Millennium Prize Problems, a list created by the Clay Mathematics Institute that offers a $1 million prize for each valid solution.
Solving questions at this level could affect fields such as physics, engineering and computer science. But it also raises a difficult question: what happens when AI produces mathematical results faster than the community can verify, understand and publish them.
OpenAI acknowledges that using open problems to evaluate new AI systems can have negative consequences. For example, a model could produce answers that are difficult to verify, change how researchersâ work is evaluated or concentrate in a few companies the tools capable of shaping the future of a discipline.
An independent group to review the results
To address these concerns, OpenAI is announcing an independent advisory group made up of mathematicians from different institutions. Its role will be to act as a link between the company, the mathematical community and the public.
The advisory group will advise on:
- The review and communication of new results.
- The mathematical importance of discoveries.
- Coordinating their dissemination.
- Academic and professional research standards.
- The use of AI tools to support research and learning.
The company says the group will not be responsible for deciding how quickly its internal mathematics research advances. Its members will not be paid by OpenAI either, and they will be able to offer unsolicited criticism, make their advice public and change the groupâs composition when they consider it necessary.
Its initial members include François Charles, Camillo De Lellis, Timothy Gowers, Martin Hairer, Nikhil Srivastava, Ulrike Tillmann, Ravi Vakil, Edward Witten and Melanie Matchett Wood. The group will be hosted at the Institute for Advanced Study.
What changes for you
This does not yet mean you can use a public tool to solve mathematics problems at this level. OpenAI is referring to an internal model and results that still need to be evaluated by specialists. The announcement also does not explain how many of the more than 100 problems have complete proofs or whether other mathematicians have validated those solutions.
What does change is the debate over how AI enters a discipline where an answer is not enough: you have to prove it, review it and understand its consequences. OpenAI wants to put more capable systems in mathematiciansâ hands, but it acknowledges that the community must help set the rules for that process.
The next key point will be public evidence. If the results withstand independent review, the question will no longer be only how much AI can solve, but how to integrate its discoveries without losing rigor, traceability or academic oversight.