The AI revolution in math has arrived

A jolting summer

It has been reported that the tipping point came in July 2025, when several AI models solved five of six problems at the International Mathematical Olympiad. Mathematicians were stunned — and rightly so. Olympiad problems are fiendish puzzles with known answers, not open research questions, but the result still felt like a wake-up call. Did a wrench just get thrown into a centuries-old workshop? For many, the answer was: yes — and that realization carried a mix of awe and unease.

From puzzles to proofs

What surprised practitioners even more was what came next. It has been reported that models not only handled contest-style problems but began helping to formulate and prove new results, sometimes producing proofs with minimal human intervention. Large language models such as ChatGPT, Claude and Gemini have become playmates and partners: tossing out ideas, suggesting strategies, finding lemmas. “This guy’s got a shovel. This guy’s got a pickax. Together we can bore a tunnel,” Terence Tao said, capturing the pragmatic optimism: lots of trial and error, and suddenly things stick. Some of these outputs now rival results published in professional journals — not every day, but often enough to change workflows.

Culture shock and the long view

Not everyone is ready to celebrate. Akshay Venkatesh warned that powerful tools risk eroding direct experience with mathematical understanding — “valuable things in our culture which we should try to keep,” he said. At the same time, talent is drifting toward industry: researchers are leaving academia for OpenAI, Google and math-focused startups like Harmonic, Logical Intelligence, Axiom Math and Math Inc. Jeremy Avigad sees why: the hunt for general intelligence is marrying machine-learning insight with mathematical precision. The emotional core here is obvious — excitement tinged with loss. No one says AI will replace mathematicians overnight, but the craft of doing math is already being refashioned, and institutions will have to catch up.

Sources: quantamagazine.org, Hacker News