The Velocity Gap

An AI solved an 80-year-old math problem that had defeated generations of mathematicians. The deeper story is what happened next — and what the institutions around it could not.

Something fundamental changed in the third week of May 2026. For the first time, a general-purpose AI reasoning model autonomously solved one of the most famous open problems in discrete geometry — a question posed by Paul Erdős in 1946 that had resisted 80 years of human effort.

The model didn’t retrieve an answer. It discovered new mathematical territory by reaching across algebraic number theory — a field not traditionally adjacent to the problem — and used ideas from that domain to construct point configurations no human mathematician had previously considered.

Fields Medalist Tim Gowers and other leading mathematicians who reviewed the work noted they were surprised not only by the solution, but by the path the model took — one humans had systematically overlooked.