AI Is Starting to Move the Erdős Frontier
-

Since late December, 15 Erdős problems have shifted from “open” to “solved,” with 11 explicitly crediting AI systems as part of the solution process. Researchers report that OpenAI’s GPT-5.2, alongside models like Google’s AlphaEvolve, has shown a noticeable jump in high-level mathematical reasoning — especially when paired with formal proof tools.
Renowned mathematician Terence Tao has taken a measured but optimistic view, noting that AI systems may be particularly effective at tackling the “long tail” of obscure Erdős problems that often have straightforward but overlooked solutions. Tools like Lean and Harmonic’s Aristotle are accelerating proof formalization, making AI-generated reasoning easier to verify and extend.
While AI isn’t replacing mathematicians, it’s increasingly acting as a powerful collaborator — one that’s beginning to push the boundaries of what problems get solved, and how quickly.