OpenAI Model Finds Counterexample to 80-Year-Old Erdős Conjecture

Researchers working in collaboration with an OpenAI model have discovered a counterexample to an Erdős conjecture that had stood for approximately 80 years. The story was shared by researchers on the OpenAI Podcast, framing the result as an early example of model-mathematician collaboration producing genuinely novel mathematical results rather than verifying known proofs.

Why It Matters

Finding a counterexample to a long-standing conjecture is categorically different from solving a benchmark task — it represents a model contributing to the frontier of human mathematical knowledge, a capability class that has not previously been demonstrated at this level.