OpenAI’s 722 math papers ignite a fight over discovery’s rules
OpenAI’s 722 math papers ignite a fight over discovery’s rules
The argument began before the latest papers landed. In late September, the Advisory Group on Mathematics and Artificial Intelligence urged labs to publish results promptly through established academic channels, disclose models, prompts and computing costs, and avoid using discoveries as promotional material. It warned that turning mathematical releases into marketing vehicles could inflict “significant harm” on the community.
That advice came after a turbulent year in which OpenAI and other labs announced results on longstanding problems faster than many researchers expected. The achievements have been real enough to command attention, but the manner of their release has bred distrust: mathematicians have questioned whether small advisory panels can represent the field, whether labs will follow their guidance, and whether AI results have drawn too heavily on human work. One critic described the company’s approach as driven less by advancing mathematics than by a desire “to win.”
On Tuesday, OpenAI released 722 manuscripts spanning 372 families of results, saying an unreleased frontier model had tackled hundreds of open questions. The company published the material in a GitHub repository, alongside revision and citation protocols, and said the average result used the equivalent of three hours of ChatGPT Pro reasoning. Greg Brockman cast the ambition in expansive terms: “towards acceleration of scientific discovery and improving quality of life for everyone.”
The response has split along a more fundamental line: what counts as valuable mathematics? Stephen Wolfram cautioned that producing formal results is not the same as producing consequential ideas. “You can make a trillion theorems easily,” he said. “The problem is most of those theorems are not ones that anybody will care about.”
Others see a reordering rather than a collapse of the discipline. Yann LeCun argued that automation of formal proof could shift mathematicians toward “new concepts, new abstractions, new definitions.” That is the promise—and the pressure—behind OpenAI’s blitz: a field built on slow validation is being asked to absorb industrial-scale output without surrendering its own standards.
Write a comment