TechCrunch Industry News AI models are starting to crack high-level math problems
Jan 15, 2026
Recent advancements in AI, particularly with GPT 5.2, are revolutionizing high-level mathematics. Engineers have used these models to solve complex problems, invoking classic theorems along the way. The rise of AI tools in math research is making formalization and proof verification more accessible. Notably, researchers focus on tackling Erdos's conjectures, pushing boundaries in problem-solving. As these tools gain traction among top mathematicians, the future of mathematical research looks increasingly collaborative with AI.
AI Snips
Chapters
Transcript
Episode notes
Engineer Finds Unexpected Full Proof
- Neil Samani tested GPT by pacing a hard math problem into ChatGPT and letting it think for 15 minutes.
- He returned to find a full, checkable solution that he formalized with Harmonic.
LLMs Can Use High-Level Math Theorems
- GPT's chain-of-thought invoked advanced theorems like Legendre's formula and Bertrand's postulate.
- The model discovered and built on prior research to produce a more complete solution than earlier work.
GPT 5.2 Pushed The Mathematical Frontier
- Since GPT 5.2's release, solved-problem volume rose and many Erdos problems got progress credited to AI.
- Terence Tao notes AI made autonomous progress on several problems and aided others by locating prior work.