OpenAI's AI model disproves 80-year-old Erdős unit distance conjecture

In May, OpenAI revealed that an internal AI model had disproved the Erdős unit distance conjecture, a problem posed by prolific mathematician Paul Erdős in 1946 that had stumped humans for 80 years. The conjecture asks: given n points in a plane, what's the maximum number of pairs that can be exactly one unit apart? Erdős provided upper and lower bounds, but a full proof remained elusive. OpenAI's model automatically searched across mathematical subfields and combined existing ideas to construct a proof. The result was shared with several top mathematicians, including Fields Medalist Tim Gowers, who called it a milestone. University of Toronto's Daniel Litt noted it's the first AI-produced result he finds inherently exciting, not just as a leading indicator.

However, the AI did not invent radically new techniques; it masterfully repurposed known approaches. Human mathematicians have since refined and extended the proof. This suggests a near-term future where AI and humans collaborate—AI bringing vast knowledge of prior work and relentless patience for tedious proof strategies, while humans contribute deeper insight. Yet the rapid pace of AI progress in mathematics raises questions about the long-term role of human mathematicians. The breakthrough marks a tangible advance: for the first time, an AI system autonomously resolved a major open conjecture, demonstrating that AI can now meaningfully contribute to pure mathematical research.