AI Solves Problems Open Since 1946 — and Mathematicians Are Rattled

In recent months, AI systems from OpenAI, Anthropic, and Epoch AI have closed several mathematical problems that had stood for decades: the Unit-Distance Conjecture (unsolved since 1946), a series of Erdős problems, and twice cracked FrontierMath — a benchmark designed specifically so that AI couldn't solve it. More than 3,000 mathematicians have signed the Leiden Declaration, expressing alarm about the profession's future.

1What Was Actually Solved

OpenAI disproved the Unit-Distance Conjecture, which had resisted mathematicians since 1946. A Carnegie Mellon team combined SAT solvers with AI-generated code to solve a Ramsey theory problem. Epoch AI twice cracked FrontierMath — a set of olympiad- and research-level problems designed so that models couldn't solve them by brute force.

2The Community Reacts: From Optimism to Alarm

Terence Tao compares AI to "a diligent and tireless grad student" and sees a golden age of collaboration ahead. Timothy Gowers, by contrast, warns of the erosion of mathematical culture: if a machine produces the proof, there is no one left to pass on the ability to appreciate its beauty. The six Millennium Prize Problems, worth $1M each, have yet to be solved by any AI.

Mathematics has become the first "pure" science in which machines are systematically outperforming the best human minds on their own turf. This changes not just the tools researchers use, but the very question of what it means to understand mathematics — and whether the profession is needed in its current form.