Why Erdős's puzzles keep falling to AI — and what mathematicians say it means
Quanta Magazine charts how erdosproblems.com — a hobbyist catalogue of the 1,200-plus problems Paul Erdős left behind — became the arena where AI labs now test their machine mathematicians. OpenAI models cracked the 80-year-old unit distance problem in May and three more Erdős problems on August 1; Google DeepMind has solved four, recovered nine forgotten solutions, and autonomously closed 9 of 353 formally posed problems. The site now counts 565 problems solved against 652 still open, and Erdős problems concentrate in number theory and combinatorics — exactly where language models are strongest.
Why it mattersThese are real, previously unsolved questions, not benchmarks, so each one that falls is a genuine capability milestone. And the reaction splits mathematics: one Princeton veteran says a problem 'loses its point' once AI can do it, while the site's founder warns of machine-written proofs no human has read — a preview of arguments every research field will soon have.
▶ Related video: The truth behind AI “solving” the Erdos problems
Read in the app — free, in 9 languages
Related stories
Twenty tries at the same task cut the best agent's score by a third.2026-10-06A model learned to write without the method that trains every AI.
2026-10-06The US got 16 countries to put AI at the centre of science.
2026-10-05Mathematicians cracked five open problems using an ordinary chat box.
2026-10-05Untuned models solved agent tasks their polished versions could not.
2026-10-04