OpenAI AI Disproves 80-Year-Old Erdős Geometry Conjecture

An internal OpenAI model has disproved the planar unit distance problem, a classic geometry puzzle that has remained unsolved for 80 years [1]. First posed by Hungarian mathematician Paul Erdős in 1946, the conjecture asked for the maximum number of pairs of dots that could be exactly the same distance apart on a flat surface [1]. While Erdős theorized that this number grew only slightly faster than the number of dots, the AI’s proof demonstrates that dots can be packed to create significantly more equal distances than previously believed possible [1].

The breakthrough occurred after OpenAI researchers prompted a large language model to investigate whether Erdős was wrong [1]. The AI bypassed standard approaches by applying complex number systems and geometry branches to generate hundreds of pages of calculations [1]. While the software produced the core discovery, human mathematicians later reviewed the work, rewrote the findings, and verified the proof [1]. The final theorem establishes that there exists an absolute constant and infinitely many integers for which the unit distance count exceeds the previously accepted limits [1].

This result marks a significant shift in mathematical research, as it is arguably the first time an AI has autonomously resolved a major open conjecture [2]. Fields Medal winner Tim Gowers described the solution as a milestone in AI mathematics, while University of Toronto professor Daniel Litt noted it is the first AI-produced result he finds exciting in its own right rather than as a mere indicator of future potential [2]. Despite the achievement, experts emphasize that the AI did not invent entirely new mathematical techniques; instead, it synthesized existing ideas and performed the tedious, high-volume calculations that human researchers often find prohibitive [2].

The success of this project highlights a growing trend of using machine intelligence to tackle long-standing problems that have resisted human effort for decades [3]. Researchers are now increasingly turning to AI to hunt for solutions to even more complex challenges, such as the Millennium Problems, which include the elusive quest to understand fluid dynamics through the Navier-Stokes equations [3].

Whether AI will eventually crack these million-dollar puzzles remains an open question, but the ability to simulate and verify complex systems has already begun to change how mathematicians approach the limits of reality. The central question now is not whether machines can contribute to the field, but how far they can push the boundaries of human comprehension before reaching a point where the logic becomes too opaque to interpret.