OpenAI Releases 700 Papers, Sparking Debate on AI in Math

OpenAI released more than 700 research papers on October 6. Some of these documents were described as atrociously written and violating community expectations of attribution. Additionally, some of the released papers were wrong. Hugo Duminil-Copin lamented that no major open problem he mentioned was left untouched by the announcement. He noted that everything was claimed to be proved in the dump, but not all at once or with such nonchalance as expected.

The classification of finite simple groups was announced by Dan Gorenstein in 1983. The case of quasithin groups was not completed for another two decades after that date. The classification cooled interest in the subject after its announcement. John Conway's favorite question is why the Monster group exists.

Frank Calegari wrote a guest post arguing that mathematicians should adapt to AI rather than be sad about it. The author believes proofs of major conjectures using AI will remain part of new mathematics. They suggest that choosing worthwhile problems may matter more than producing proofs in the future. Senior mathematicians are encouraged to create future problems and explore new ideas rather than simply using AI to answer existing questions. The author proposes that the most effective tools to affect how people act in mathematics will be economic rather than rhetorical. They suggest evaluating all work as if it is AI-assisted, making hiring much more about personal interactions rather than a paper file, and demanding more from journals and abandoning those that do not meet standards.