OpenAI releases hundreds of math results, mathematicians skeptical

OpenAI has released hundreds of results on major open questions in mathematics and theoretical computer science, Scientific American reports. The results were posted on a GitHub repository at 6 p.m. EDT. Many have already been verified in Lean, a programming language that checks proofs, so they are almost certainly correct. Mathematicians will need months to fully understand them.

The new results come from the same model that produced OpenAI's earlier solution to the Navier-Stokes problem. That earlier work used a swarm of 10,000 AI agents and cost millions of dollars. This time, an OpenAI spokesperson told Scientific American that nearly every result came from a single prompt to a single AI agent. If true, it suggests that powerful mathematical ability could soon be accessible to anyone. But skeptical mathematicians point to OpenAI's history of bold claims and limited transparency.

"Until and unless they release the model and people can replicate their results, I think you should treat any claims about one-shotting problems with a single agent as unverified. We should ask for receipts."

Andrew Sutherland, a mathematician at MIT, told Scientific American. The spokesperson added that some results might have taken multiple attempts.

The controversy over OpenAI's handling of the Navier-Stokes solution led the company to announce an independent advisory group of mathematicians on September 21. The group recommended that any company releasing AI-generated results should make public the model, exact prompt, and compute time. OpenAI has not followed those recommendations fully: it has released only the average compute time and some statistics, not the prompts. The spokesperson said the company takes the guidelines seriously but is not bound by them. The group specifically criticized the use of proprietary models that mathematicians cannot access. OpenAI says it is working to release the model as soon as possible.

The debate has divided mathematicians. Daniel Litt of the University of Toronto told Scientific American that secrecy serves no purpose.

"If we want to know the answers to these math questions, I see no reason why we should ask the company to keep them secret from us. To me, it's going to be a good thing for mathematics."

But Terence Tao has criticized the pace of AI-generated results, calling it "insane."

The OpenAI spokesperson told Scientific American that many of the new results are not yet understood even by the company's own mathematicians, but that OpenAI has no plans to slow down. The company sees the math problems as a critical test of its AI's progress. For mathematicians who value truth as an end, the consequences of this fast-moving race are becoming harder to predict, Scientific American reports.

• AI transparency
• Lean
• mathematicians
• mathematics
• Navier-Stokes
• OpenAI