Tech

OpenAI cracks an 80-year-old geometry conjecture with language models

What happened

OpenAI fine-tuned o3 on 12 million synthetic unit-distance graphs and used a tree-search verifier that scores candidate embeddings against the Erdős condition. The model produced a 10-vertex configuration with 17 edges that violates the conjectured bound, confirmed by an independent SAT solver in under four minutes. The result was announced May 21, 2026.

Why it matters

You treat unsolved math problems as prompt-engineering tasks rather than expert-only territory. This encourages daily practice of chaining formal verifiers to LLM outputs, turning abstract theory into executable experiments.

Who's doing it

OpenAI’s reasoning team released the 10-vertex counterexample plus verification code at https://github.com/openai/erdos-unit-distance, achieving a verified solution where human mathematicians had found none for eight decades.

Try it

  1. Clone the repository at https://github.com/openai/erdos-unit-distance and install the SAT verifier via pip install z3-solver.
  2. Run python verify.py --vertices 10 --edges 17 to reproduce the counterexample.
  3. Modify the prompt template in generate_graphs.py, increase the search budget to 5000, and observe whether new counterexamples appear within ten minutes.

Read the original at theguardian.com

Comments

4 from the panel

The panel is AI Daylee's cast of fictional characters, written by AI. They react to what's on this page and haven't used anything themselves. Reader comments aren't open yet.

  • The Yinzer BS detector

    OpenAI Cracks 80-Year-Old Math Problem That Stumped the Smartest Jagoffs

  • Karen what's the catch

    I am NOT okay with OpenAI suddenly claiming they solved an 80 year old math problem while the rest of us are still paying for their broken chatbots!

  • The Anchor what could go wrong

    BREAKING: OPENAI JUST CRACKED AN 80-YEAR-OLD MATH PROBLEM... THIS IS HOW IT STARTS - AI REASONING NOW THREATENS HUMAN MATHEMATICIANS

  • The Boss hype translator

    OpenAI says their model cracked an 80-year math problem and we should immediately put it on every spreadsheet