Altman Says OpenAI Will Keep Releasing AI-Generated Math Proofs Despite Mathematicians' Complaints
Sam Altman has said OpenAI does not intend to stop releasing its AI-generated mathematical results, even though some mathematicians say they are struggling to make sense of them. He made the remarks in an episode of the Cheeky Pint podcast released on Thursday, October 29.
Altman said he sympathizes with the mathematicians. He also argued that the public's right to new scientific knowledge outweighs the disruption the releases cause. In his view, humanity deserves fundamental knowledge, and OpenAI should not sit on a major fundamental science discovery. The Verge reported these remarks as the core of the interview.
Asked whether OpenAI has more mathematical proofs beyond the hundreds already released, Altman answered "We do. Of course." He also acknowledged that the company's pursuit of mathematical problems "got out of control" for a short period. He described the effort as deliberate, aimed at probing the limits of its AI models.
The tension has been building for weeks. In early October, WIRED reported that one mathematician perceived "mobster behavior" from leading AI companies. Altman later said that mathematicians had gone through phases after OpenAI began its push into the field, in remarks reported by Yahoo Tech. OpenAI has also promoted its work in this area on its own podcast. One episode features reasoning researchers Alexander Wei, Hongxun Wu, and Lijie Chen discussing an AI finding that mathematicians had missed for decades.
The broader context here is verification rather than novelty. A proof is a formal chain of logic showing that a claim must be true, and a model-generated proof is only as useful as the community's ability to check it. A steady flow of results puts that checking capacity under pressure. Mathematics has a long tradition of peer review, and a result that specialists cannot read, understand, and confirm remains a claim. The mathematicians' complaints are, at heart, about throughput and legibility, not about whether the work is interesting.
In my view, the most useful test for these releases will be the one mathematicians are already applying. Fundamental results that survive expert scrutiny, and that other researchers can build on, would make Altman's case far more concrete than any podcast argument. Results that arrive faster than the field can evaluate them will generate headlines without necessarily generating knowledge. The gap between those two outcomes is where the next few months will be decided.
OpenAI has released hundreds of AI-generated proofs, says it holds more, and its chief executive has committed to continuing the releases despite the friction they cause in the mathematical community.
