tech
January 14, 2026
AI models are starting to crack high-level math problems
Since the release of GPT 5.2, AI tools have become inescapable in high-level mathematics.
TL;DR
- Neel Somani, a software engineer, found that OpenAI's GPT 5.2 model can solve complex mathematical problems.
- The AI model demonstrated impressive mathematical reasoning, citing axioms and finding solutions to problems posed by mathematicians like Paul Erdős.
- Since Christmas, 15 Erdős problems have been marked as solved, with 11 crediting AI involvement.
- Terence Tao noted that AI is making meaningful autonomous progress on Erdős problems and may solve easier ones more efficiently than humans.
- Tools like Harmonic's Aristotle and the open-source proof assistant Lean are aiding in the formalization of mathematical proofs.
- Mathematicians and computer science professors are increasingly using AI tools, lending credibility to their capabilities.
Continue reading
the original article