OpenAI Claims Progress on an 80 Year Old Geometry Puzzle
OpenAI applied large language models to the Paul Erdős planar unit distance problem. The models generated new upper bounds on the chromatic number of the plane. The work demonstrates measurable gains in multi step mathematical reasoning without external symbolic solvers.
You see that current models can now iterate on open research questions rather than only regurgitate textbook proofs. This shifts your workflow from verification to guided exploration when you face unsolved problems in math or logic.
OpenAI published the findings on 21 May 2026 and reported that their system improved the best known upper bound from 7 to 6 in one configuration. The result was verified by independent mathematicians before release.
Step 1: Go to https://www.theguardian.com/technology/2026/may/21/openai-paul-erdos-maths-problem-breakthrough and read the problem statement. Step 2: Copy the formal problem into ChatGPT o1 and ask it to propose candidate colorings or graph constructions. Step 3: Export the generated graphs to Python with NetworkX and run a brute force chromatic number check to validate or refute each candidate.