OpenAI Publishes 719 Math Papers by an Unreleased AI Model

Admiration and unease

The release follows OpenAI's September announcement of a solution to the Navier-Stokes problem, one of mathematics' great challenges. That work drew criticism from mathematician Tristan Buckmaster, who questioned how much earlier research contributed to the AI's result. This time, the company says it consulted an independent group of experts to improve how it shares discoveries.

Steven Kelk, associate professor of theoretical computer science at Maastricht University, told Dutch newspaper Volkskrant that he views the results with admiration. In his view, the AI may have delivered progress equivalent to ten or twenty years of research in one go. Yet he also described a «great sense of emptiness». For Kelk, research means making mistakes, collaborating and discovering new ideas along the way, and part of that experience disappears when a machine simply hands over the answer.

Terence Tao has also voiced concern about researchers who use AI without sufficiently understanding the results. The independent advisory group AGMAI has asked that such advances be shared transparently and help build human knowledge.

Funding for understanding

OpenAI announced a $5 million contribution to a Simons Foundation initiative that aims to help mathematicians study and understand AI-generated proofs. The effort suggests that checking and interpreting machine-made results may become as important as producing them.