AI math bot stuns world’s top minds at secret Berkeley showdown

At a hush-hush Berkeley meetup, 30 elite mathematicians tried to stump OpenAI’s o4-mini with unsolved math problems and… it mostly failed. The AI cracked complex challenges in minutes, showing reasoning skills one scientist called “frightening.”

‍

It even got cheeky: “No citation necessary because the mystery number was computed by me!” Despite offering $7,500 bounties for every unsolved problem, the humans struggled to outwit it. One expert likened the AI to a top-tier grad student—only faster, sharper, and possibly tireless. Mathematicians left stunned, shaken, and very much outpaced.