The World, the Universe and Us Mathematics is Undergoing the Biggest Change in its History
24 snips
Mar 13, 2026 
Alex Wilkins, New Scientist reporter covering AI and mathematics, explains how AI is now cracking competition problems and research puzzles. He discusses why mathematicians feel threatened, how specialised models differ from chatbots, and the push to formalise proofs so machines can check them. The conversation also recalls AlphaGo’s creative legacy and what rapid AI progress means for the field.
AI Snips
Chapters
Transcript
Episode notes
Secret Ten Problem Test Revealed AI Research-Level Ability
- Mathematicians built a secret 10-problem test set; OpenAI solved five and DeepMind six, surprising experts.
- The test used problems withheld from training to rigorously evaluate AI on real research-level questions.
Scaffolds Let Math AIs Explore And Self Check
- Advanced math AIs are built from large language models plus scaffolding that explores multiple solution paths and self-checks.
- Alex Wilkins explains these systems run parallel attempts and pit solutions against each other, unlike public ChatGPT.
Unlimited Compute Enables Systematic Math Experimentation
- AI offers effectively unlimited compute to test thousands of variant mathematical problems and compare methods systematically.
- Terence Tao told Alex Wilkins AI can let mathematicians run controlled comparisons like randomized trials across many problem instances.