Overview
- OpenAI says a new general‑purpose reasoning model produced an original proof for Erdős's 1946 unit‑distance problem that establishes a superlinear lower bound u(n) ≥ n^(1+δ) with δ>0.
- The company published a 125‑page distilled account of the argument and cited corroborating statements from noted mathematicians including Noga Alon, Melanie Wood, Thomas Bloom and Timothy Gowers.
- OpenAI has not yet released the underlying model or the full, unabridged chain of AI reasoning, and mathematicians are urging access to complete materials for independent verification.
- According to reports, human mathematicians took the AI's construction and immediately improved the lower bound, showing collaborative follow‑up work began as soon as the idea circulated.
- If validated, the result would overturn an eight‑decade consensus that unit‑distance pairs grow roughly linearly and could change approaches in geometry and related fields, but reproducibility and transparent review are the next critical steps.