OpenAI Floods Math With 400 AI Results as 42% of 719 Manuscripts Are Formally Verified
Updated
Updated · The Verge · Oct 9
OpenAI Floods Math With 400 AI Results as 42% of 719 Manuscripts Are Formally Verified
3 articles · Updated · The Verge · Oct 9
Summary
Nearly 400 AI-generated results across 719 manuscripts have left mathematicians scrambling, with many saying it could take years to determine which claims are correct, important, or already known.
Only 300 top-line results—about 42% of the manuscripts—were formalized in Lean, and OpenAI had already revised more than a dozen papers and removed three over a sign error.
Researchers said some papers appear genuinely elite, with “tens” of results touching major targets such as the Riemann hypothesis, Hodge conjecture and 4D Kakeya, but complained that many write-ups were hard to follow and thinly attributed.
That mix of caliber and uncertainty is already disrupting careers: mathematicians told The Verge grant proposals and entire research programs were effectively wiped out, especially in probability, combinatorics and theoretical computer science.
OpenAI says it will fund workshops and keep evaluating frontier models on math, while critics argue AI labs are offloading verification and explanation onto academia faster than the field can absorb.