Back to Dwarkesh Patel

Grant Sanderson (@3blue1brown) – AI and the future of math

Dwarkesh PatelJune 30, 20261h 33m
Topics63
AI Progress in Mathematics as Indicator for Broader Capabilities0:00IMO Performance and Domain Differences1:03Millennium Prize Problems and Cross-Domain Connections3:06Contrasting Mathematical Discovery with White-Collar Automation5:30Goalpost Movement and Next Benchmarks7:31Conjecture and Definition Generation9:03Measuring Conjecture-Generating Ability10:30Training Limitations and Verification Loops11:34Galois Theory Example13:02Delayed Recognition of Group Theory21:03Rewarding Conceptual Elegance in AI23:12Three Paths to Solving Major Conjectures23:34Kolmogorov Complexity and Elegance24:07The Goal of Understanding24:31Worries About Mathematics and AI Understanding26:01The Unit Distance Conjecture Counterexample27:00Three Ways of Solving the Riemann Hypothesis27:30Lightning Bolts Between Fields28:02Mountain Building and Alien Mathematics29:00The Fall of the Theorem Economy29:34Proof Versus Understanding31:03Unsolved Expository Problems32:03The Distinction Between Proof and Explanation33:04Exposition as Explanation33:31Novel Insights and Clear Communication34:00AI's Role in Explanation35:02The Future Role of Mathematicians36:02Curation Over Creation37:32Connecting Ideas as the Core of Breakthroughs38:31The Langlands Program39:02AI as Supercharged Connectors40:30Why the Connection Unlock Hasn't Happened Yet41:33Questioning the Premise of Token Generation43:30Parallelization Advantages of AI45:32Quantity Has a Quality of Its Own46:32Escaping Context in Mathematical Problem-Solving48:04The IMO "Troll Problem" and Context Escape48:32Systematic Context Refreshing as an AI Advantage50:02Einstein's Heuristics and Multiple Research Programs51:06Cursor and Practical AI Applications52:33Verifiability vs. Grindability in AI Progress54:05Sample Efficiency and Containerization55:04Lean, Formalization, and Process-Based Supervision56:31Endless Exploration with Lean and Mathlib57:36Lean vs. Natural Language for Automated Research59:30Exploring Axiom Spaces and Terry Tao's Project1:01:32Natural Language Verification and Meta-Verification1:03:04LLM Judgment Capabilities and Writing Limitations1:04:35The Value of Formal Verification for Generated Content1:05:38Extending Human Knowledge Corpora1:07:02Why Writing Progress Lags Behind Code and Math1:07:34LLM Explanations vs. Human Writing Quality1:09:00Autoregression and the Challenge of Insightful Writing1:10:31Teaching LLMs Spaced-Repetition Prompts and Mentalizing1:11:34Botox and Theory of Mind1:12:57Jane Street Culture1:15:00Using LLMs for Learning1:16:08Optimal Learning Workflow1:19:31Limitations of LLMs as Explainers1:21:02Career Advice for Mathematics Students1:22:31Future Demand for Mathematical Distillation1:28:04Practical Applications of Accelerated Mathematics1:29:33Potential Disconnect Between Mathematical Progress and Utility1:33:02
In a Nutshell

Grant Sanderson argues that AI's rapid progress in math—driven by verifiable, grindable domains like IMO problems and Lean formalization—signals broader capabilities, but the hardest parts to automate are generating novel conjectures, definitions, and insightful cross-domain connections rather than raw theorem-proving. He highlights that AI's advantages include parallelization, systematic context-escaping, and endless exploration of axiom spaces or Mathlib extensions, yet warns that "mountain-building" new theories or distilling compressed, human-understandable insights may lag behind brute-force solutions. Ultimately, mathematicians' roles will shift toward curation—deciding what ideas are worth pursuing—while demand grows for humans who translate AI output into clear, motivating explanations and identify physically relevant results.

AI-Generated Notes

These notes were generated by AI and may contain inaccuracies.

Grant Sanderson runs 3Blue1Brown and is documenting AI progress in mathematics. AI has made the fastest progress in mathematics compared to any other field, and developments here will indicate what happens elsewhere as AI improves. Three years prior, when asked whether AIs achieving gold at the International Math Olympiad would constitute AGI, Sanderson responded that it would be another benchmark like others AI systems pass, without a singular "aha" moment. AI has improved generally since then, but the prediction held.

The IMO contains four categories: geometry, number theory, algebra, and combinatorics. AI systems are very good at geometry, solving problems in nineteen seconds via brute force since 2024. For students, geometry also has a brute force approach. Combinatorics problems are more playful and puzzly. In the year discussed, two combinatorics problems appeared on the test. If more geometry questions had been present, the AI would have achieved gold. The system struggles on combinatorics problems, which some view as requiring more creativity.

Solving a Millennium Prize problem like the Riemann hypothesis may not equate to automating white-collar work. AI systems excel at deep knowledge within specific domains but have not consistently found lightning-bolt connections between domains. Sparks of cross-domain connection-finding are emerging. The Riemann hypothesis solution might involve such bridging, as in the story of Hugh Montgomery and Freeman Dyson at the IAS. Montgomery studied statistical correlations between pairs of zeros of the Riemann zeta function and derived a formula resembling one over sine squared. Dyson recognized this expression from eigenvalues of random Hermitian matrices in nuclear physics, prompting exploration of random matrix theory's relevance to the zeta function.

Sign in to read the full notes

Get access to AI-generated notes, topic timestamps, and more.