Back to news
researchOpenAI2026-10-30

Altman says he understands mathematicians' position but OpenAI will not stop, with math research spending in the hundreds of millions

OpenAI CEO Sam Altman said on the Cheeky Pint podcast that he understands mathematicians' position as AI research results proliferate, but OpenAI needs to keep publishing scientifically valuable discoveries.

OpenAI CEO Sam Altman said on the Cheeky Pint podcast that he understands mathematicians' position as AI research results proliferate, but OpenAI needs to keep publishing scientifically valuable discoveries. He believes people have the right to know the basic laws governing the universe and the company should not keep major fundamental discoveries internal. Altman acknowledged that how OpenAI had published mathematical results needed improvement, and said he hopes to advance science while respecting the research communities that have driven mathematics for centuries and helping researchers adapt to AI. OpenAI has published a large volume of AI-driven mathematics results this year, triggering disputes over publication practice, proof quality and academic attribution, with researchers criticizing some papers for lacking background explanations, references and clear argumentation, which forces mathematicians to spend substantial time verifying results. Altman said OpenAI deliberately pushed mathematics research hard for a short period to probe model limits, revealing an internal Slack channel called Model Name-Hold My Beer where staff submitted the hardest problems from their academic careers, with results exceeding expectations. He also admitted feeling a stronger-than-expected sense of loss at AI solving problems once considered extremely difficult, since part of the value of mathematical research is that researchers may solve the hardest questions through long effort. He estimated OpenAI has spent hundreds of millions of dollars on mathematical problem research, with work on Navier-Stokes equations costing about 10 million dollars and possibly more, and that with better models and methods similar work might cost about one percent of current levels in future.

OpenAI奥尔特曼数学研究学术归属NS 方程科学发现