Altman Admits AI-Generated Math Outputs Are Overwhelming Mathematicians, But Has No Plans to Hit the Brakes

OpenAI CEO Sam Altman talked about the trouble his company's AI-generated mathematical results have brought to mathematicians on the Cheeky Pint podcast released Thursday. The Verge's report notes that over the past few months, a flood of AI-generated outputs has poured in to mathematical researchers, forcing many of them to spend extra time sorting out which results are worth reviewing and which are just meaningless noise.

Altman's exact quote goes: "I think, generally speaking, the world should get to understand how the universe works." He then added: "But I have sympathy for the process." The "process" here refers to the work mathematicians do to digest and verify these AI-generated outputs. His stance is unambiguous: he recognizes the existing hassle, but made no promise to slow down the output pace. The report also mentions he hinted that OpenAI's stockpile of findings is far larger than what has been released, and more is coming later.

The background behind this issue is straightforward. As the report describes, the speed of AI-generated outputs has created massive pressure on the verification end. The field of mathematics relies heavily on rigorous proofs and peer review. Even a seemingly elegant conjecture still requires someone to break down the full proof and check every single step. When the generation side accelerates sharply while the verification side does not speed up at the same rate, a massive backlog forms. The new extra task for mathematicians could very well be acting as reviewers for AI outputs.

This is probably the real implication of Altman's expression of sympathy: he sympathizes with the process of verification, but will keep rolling out new results all the same. For mathematicians, the annoying part is that AI does not get tired, but humans do. For ordinary users, this also explains why AI's mathematical capabilities are improving so quickly, while the outcomes fully digested and recognized by the academic community always lag far behind.

Podcast video: this episode of Cheeky Pint

Original article: The Verge