Altman says OpenAI will keep flooding mathematics with AI results despite pushback

OpenAI has no plans to slow its release of AI-generated mathematical results, despite months of complaints from mathematicians who say the flood has left them scrambling to verify work they had no part in producing. CEO Sam Altman, speaking on the Cheeky Pint podcast in an episode released Thursday, said he understands the frustration but believes the payoff justifies the disruption.

"I think, on net, the world deserves to understand how the universe works," Altman said, as reported by The Verge. "But I do have empathy for the method."

A defense of the flood

According to The Verge, Altman framed the issue as a matter of principle: humanity has a claim on fundamental scientific knowledge, and an organization that finds something significant should not keep it to itself. "I feel unequivocal on this point: Humanity deserves fundamental knowledge, and we should not be sitting on a major fundamental science discovery," he said. He did acknowledge that OpenAI needs to get better at how it communicates what it finds.

The comments are notable because, per The Verge, they represent one of the first times Altman has directly addressed his company's effect on mathematicians. OpenAI has released hundreds of mathematical results this month alone, the outlet reports, typically accompanied by little more than brief writeups that researchers have described as poorly written and hard to follow.

Criticism on two fronts

Mathematicians who spoke to The Verge have faulted OpenAI for inadequate attribution, missing context and insufficient verification, which in practice hands academics the labor of checking and interpreting the results while the company moves on to producing more.

But OpenAI is now also drawing the opposite criticism. The Verge reports that researchers this week expressed frustration over what they see as the company's gatekeeping, following its announcement of three major claimed breakthroughs in mathematics, physics and computer science. Details of one of them, involving Yang-Mills theory, the mathematical foundation of modern particle physics, have not been made public. OpenAI attributes the delay to recommendations from a hand-picked group of experts on how best to release the work, guidance that runs counter to what an earlier group of consulted mathematicians advised and that the company describes as part of a new approach to communicating findings.

Altman also indicated that what has been published is only part of the stockpile. Asked whether OpenAI holds more than the hundreds of proofs already released, he replied: "We do. Of course."

How it happened, and what it cost

Altman acknowledged that the push to attack mathematical problems did "get out of control" for a short stretch, but said that was deliberate, a way to probe the limits of the company's AI models. He described an internal Slack channel whose name amounted to a dare, where employees threw the hardest problems they could find at the systems to see how they held up.

He also put numbers on the effort. According to The Verge, Altman estimated OpenAI has spent on the order of hundreds of millions of dollars on mathematical problems, including roughly $10 million, "maybe a little more," on its solution to Navier-Stokes, one of the famous Millennium Prize problems. That level of spending offers a glimpse of the resources AI labs are now directing at scientific questions, and it has fueled concern among researchers that they could be priced out of their own fields.

Why it matters

The standoff between OpenAI and mathematicians previews a problem that reaches well beyond one discipline. AI systems can now generate plausible research results far faster than humans can validate them, and the lab producing them controls both the pace of publication and the quality of the documentation around it. Altman's stated principle, that disseminating knowledge outweighs the disruption it causes, answers none of the practical questions: who verifies the proofs, who receives credit, and how a field absorbs hundreds of results it cannot yet check. Meanwhile, his admission that OpenAI is withholding many more discoveries, produced with budgets no university department can match, suggests the imbalance between generation and verification will widen before it narrows.