Grant Sanderson – AI and the future of math
Summary
Dwarkesh Patel interviews Grant Sanderson about AI’s rapid progress in mathematics, using it as a lens for understanding broader AI capabilities. They discuss how AI has mastered benchmarks like the IMO but still struggles with creative conjecture generation and deep theoretical unification. The conversation explores whether future breakthroughs, such as solving the Riemann Hypothesis, will yield human-understandable insights or opaque proofs, highlighting the distinction between automated verification and genuine comprehension.
Topics discussed
Introduction and the spiky frontier of AI capabilities
AI's potential in mathematics: Creativity vs. brute force
The Riemann Hypothesis and the nature of mathematical insight
History of Group Theory: Abel, Galois, and symmetry
AI-generated proofs: Elegance, verification, and understanding
Gemini 3.5 Live Translate demo and real-time translation
AI as a tool for connecting disparate mathematical fields
The role of human curators and the future of math research
Systematic exploration and multi-agent AI strategies
Lean, formal verification, and the Mathlib repository
AI limitations in writing, explanation, and theory of mind
Using LLMs to learn mathematics and navigate concepts
Future outlook: AI's impact on physics and economic progress
Listen ad-free on Castria