OpenAI says its model solved 100 open math problems, then hired nine mathematicians to talk about it
On Monday September 21, OpenAI announced an independent advisory group of nine mathematicians, hosted at Princeton's Institute for Advanced Study. The number that came with the announcement is the headline: an internal OpenAI model has resolved more than 100 open problems across mathematics, including Navier-Stokes, one of the Millennium Prize problems.
Then read the scope of the group, which is where the real information is. The advisers are explicitly not responsible for advising OpenAI on how to pace its internal progress on mathematics. They assess significance and coordinate release timing. They receive no payment, they can give unsolicited advice, and they can make their views public. In other words this is a body with a megaphone and no brake pedal, and OpenAI wrote that down rather than fudging it, which is at least honest.
The context nobody at OpenAI mentioned in the press release: earlier in September, 25 Fields Medal winners signed an open letter saying AI labs are threatening their intellectual work while trying to one-up each other with solutions to famous problems. Exactly one member of this new nine-person group, Camillo De Lellis of IAS, also signed that letter. One out of twenty-five. That ratio is the most informative fact in the entire story, and it tells you the advisory group is drawn from the part of the community that was not objecting.
IAS itself put out a line that reads like it was written by a lawyer who had seen this movie before: we do not have decision making power at any AI company, and the responsibility for the decisions made by any company will rest with that company. That is an institution accepting the venue and refusing the liability, and they are right to.
The thing to hold onto is that 100-plus is an unverified count from an internal model, and a resolved open problem means nothing until other mathematicians have read the proof. Navier-Stokes in particular will take months of adversarial reading by people who want it to be wrong. That process is the product here, not the announcement. The advisory group's one genuinely useful function is deciding what gets released and when, so the field gets proofs it can check rather than press releases it cannot.
← Back to all articles
Then read the scope of the group, which is where the real information is. The advisers are explicitly not responsible for advising OpenAI on how to pace its internal progress on mathematics. They assess significance and coordinate release timing. They receive no payment, they can give unsolicited advice, and they can make their views public. In other words this is a body with a megaphone and no brake pedal, and OpenAI wrote that down rather than fudging it, which is at least honest.
The context nobody at OpenAI mentioned in the press release: earlier in September, 25 Fields Medal winners signed an open letter saying AI labs are threatening their intellectual work while trying to one-up each other with solutions to famous problems. Exactly one member of this new nine-person group, Camillo De Lellis of IAS, also signed that letter. One out of twenty-five. That ratio is the most informative fact in the entire story, and it tells you the advisory group is drawn from the part of the community that was not objecting.
IAS itself put out a line that reads like it was written by a lawyer who had seen this movie before: we do not have decision making power at any AI company, and the responsibility for the decisions made by any company will rest with that company. That is an institution accepting the venue and refusing the liability, and they are right to.
The thing to hold onto is that 100-plus is an unverified count from an internal model, and a resolved open problem means nothing until other mathematicians have read the proof. Navier-Stokes in particular will take months of adversarial reading by people who want it to be wrong. That process is the product here, not the announcement. The advisory group's one genuinely useful function is deciding what gets released and when, so the field gets proofs it can check rather than press releases it cannot.
Comments