Skip to content

Fields Medalist: AI Can Do Every Basic PhD Problem in Days

Into the Impossible With Brian Keating

Hosted by Brian Keating

1h 6m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

Brian Keating is joined by a Fields Medalist to discuss the intersection of advanced mathematics, minimal surfaces, and artificial intelligence. They examine Jim Simons's pioneering work on minimal surfaces and singularities in higher dimensions, and how mathematical techniques developed for geometry unexpectedly apply to physical problems like ice melting in water. The conversation then turns to the rapid rise of AI in mathematics, touching on recent claims by tech companies that their models can solve complex problems and millenium challenges like Yang-Mills theory and Navier-Stokes. The guest argues that while AI can rapidly complete proofs, it often misses the deeper conceptual connections and foundational understanding that humans require, creating challenges for student training and the future of mathematical research.

  • Jim Simons proved that every minimal surface up to dimension seven is smooth, but an integral diverges in dimension eight where singularities can appear. 10:04
  • Mathematical techniques developed to study minimal surface singularities can be applied to completely different physical problems like ice melting in water. 19:58
  • AI models trained on past mathematical literature can execute proofs rapidly, but they often fail to grasp the underlying conceptual connections between different mathematical fields. 36:06
  • Protecting the intellectual training of students is crucial because bypassing foundational problem-solving through AI risks creating severe generational gaps in expertise. 46:18

Chapters

  1. 0:00
    Podcast Introduction and Advertisements

    The episode opens with a series of commercial ads before the host begins the discussion.

  2. 0:57
    OpenAI’s Yang–Mills Proof and Transparency Concerns

    The conversation shifts to OpenAI’s recent claim of a rigorous proof for the Yang–Mills existence problem and the mathematicians’ demand for full transparency.

  3. 1:36
    Mathematicians’ Perspective on AI‑Generated Proofs

    Alessio explains why mathematicians are skeptical of AI‑produced proofs and the need for expert verification.

  4. 3:17
    Future of AI in Mathematical Research

    Discussion on how AI will shape future research priorities and the role of human mathematicians.

  5. 6:00
    History of Jim Simons and Minimal Surfaces

    The host recounts Jim Simons’ story and introduces the concept of minimal surfaces and their significance in mathematics.

  6. 12:00
    From Minimal Surfaces to Free‑Boundary Problems

    Alessio discusses how techniques from minimal surface theory were applied to the melting‑ice Stefan problem and the importance of singularities.

  7. 18:00
    Mathematics, Physics, and the Role of Proofs

    The dialogue explores how proofs illuminate physical intuition and the interplay between mathematical uniqueness and physical reality.

  8. 26:00
    Educational Implications of AI Tools

    The discussion turns to how AI affects learning at all levels, from high school to PhD, and the challenges of maintaining motivation and critical thinking.

  9. 30:00
    Future of Mathematics and Society in an AI‑Driven Era

    The episode concludes with reflections on the responsibilities of academia, industry, and society to balance rapid AI progress with human development and understanding.

  10. 39:54
    AI’s Impact on Mathematical Research and Community

    The speakers debate how super‑human AI models change the pace of discovery, the risk of over‑reliance, and the need for human oversight.

  11. 50:05
    Introduction and Host Greeting

    Brian Keating opens the episode, setting the stage for a discussion on AI’s impact on mathematics.

  12. 52:06
    Defining AI Creativity in Mathematics

    The conversation turns to what constitutes creativity for AI and how it differs from human perception.

Highlights

  • 4:45

    Jim Simons Fundraising Storystory

    The guest shares an anecdote from a 2019 dinner where Jim Simons raised money for a mathematics institute by demanding billionaires write a 100,000 dollar check before leaving the room, reassuring the guest he was exempt because he was just a mathematician.

  • 10:05

    Minimal Surfaces in Dimension Eightdata

    Jim Simons found that minimal surfaces remain smooth up to dimension seven, but in dimension eight an integral diverges and singularities like the Simons cone can emerge.

  • 16:28

    Theorem Persons Versus Counterexamplesinsight

    The guest explains that mathematicians have different mindsets, separating those who enjoy building counterexamples from those like himself who prefer constructing theorems and proving what works under specific assumptions.

  • 19:58

    Surprising Math Connectionsinsight

    Techniques from a thousand-page work on minimal surface singularities were surprisingly applied to completely different problems like the ice melting water Stefan problem.

  • 36:06

    Superhuman AI Capabilitiesinsight

    AI models have reached superhuman capabilities in math by combining vast computational power and strategies across different fields, though they are trained on past human knowledge.

  • 46:18

    The Threat of AI to Trainingdebate

    The guest expresses concern that if AI can solve basic PhD problems in days, students may lose motivation to endure the struggle of learning core mathematical concepts.

  • 56:34

    Missing the Deeper Connectioninsight

    When OpenAI solved the De Georgi conjecture, the proof was specific to the Allen-Khan equation and missed the conceptual connection to minimal surfaces that originally motivated the problem.

Quotes

  • “I think, I don't know what is the rationale behind that. I haven't seen the proofs.”
    Unnamed speaker · 1:36
  • “And the astonishing fact, if you want, is the following. So many mathematicians ask themselves, okay, are these surfaces still, despite being maybe more bent and so on, will they remain smooth? Or will they eventually cross?”
    Unnamed speaker · 9:34
  • “I mean, as humans, we need the time to adapt and understand and adjust, you know?”
    Unnamed speaker · 43:54
  • “I think the goal is there. But if it's about understanding a theory, then no, it's totally useless.”
    Unnamed speaker · 1:02:10
  • “And that means we need experts even more and that in scientists even more, because if the production of, it's increasing so quickly, you need experts to understand what's happening and what to do next and what to analyze.”
    Unnamed speaker · 58:12
  • “I mean, sometimes it depends. I mean, for me, it depends on what, what the goal is, right? So if the theorem tells you, you don't know how, you don't understand it, but it gives you the formula to your cancer, who cares? I mean, then we just apply it and we build all the molecules we want, right?”
    Unnamed speaker · 1:01:57
  • Home Depot 0:00–0:29
  • Progressive 0:41–0:52
  • Home Depot 12:38–13:08
  • Home Depot 39:24–40:01
  • Home Depot 51:35–52:05
  • Home Depot 1:03:44–1:04:14

About this episode

A Fields Medalist says AI can now do every basic PhD problem in hours or days. The question is no longer whether machines can prove theorems. It is who will still understand them. Alessio Figalli won the 2018 Fields Medal and is a professor at ETH Zurich. He is one of 25 Fields Medalists who signed the declaration on AI and mathematics. We cover: why Jim Simons got stuck in dimension 8, how a minimal-surface technique cracked the melting-ice problem, what happens when 10,000 AI agents race human mathematicians, and what OpenAI's Yang-Mills proof does and does not tell us. A proof nobody understands can be worse than no proof. Recorded in late October 2026, two days after OpenAI's Yang-Mills announcement. 00:00 Why 10,000 AI agents worry a Fields Medalist 00:57 Jim Simons asks for a $100,000 check 02:43 Why minimal surfaces break in dimension 8 08:42 Why not stop at dimension 3? 12:12 Counterexample people vs theorem people 13:15 How minimal surfaces cracked melting ice 20:23 Why melting ice doesn't go fractal 21:27 One equation, infinitely many futures 26:30 Is all of mathematics secretly physics? 30:01 25 Fields Medalists vs the AI labs 33:01 The Navier-Stokes race 37:40 Would you have banned the espresso machine? 39:04 Why AI could empty the PhD pipeline 45:20 Can AI have Einstein's happy thought? 48:36 The connection OpenAI's proof missed 53:28 OpenAI's Yang-Mills proof, two days in 56:48 Is a proof nobody understands useless? Get the transcript, bonus content, and my Monday M. A. G. I. C. Message: https://briankeating.com/yt Have a .edu email and live in the USA? You automatically win a meteorite: https://BrianKeating.com/edu Subscribe: https://www.youtube.com/DrBrianKeating?sub_confirmation=1 Support Into the Impossible on Patreon: https://www.patreon.com/drbriankeating Join this channel for perks, Office Hours and your name in the Member Roster: https://www.youtube.com/channel/UCmXH_moPhfkqCk6S3b9RWuw/join My books: Losing the Nobel Prize: http://amzn.to/2sa5UpA Think Like a Nobel Prize Winner: https://a.co/d/03ezQFu Focus Like a Nobel Prize Winner: https://a.co/d/hi50U9U Galileo's Dialogue (first-ever audiobook): https://a.co/d/iZPi9Un X: https://twitter.com/BrianKeating Substack: https://briankeating.substack.com Blog: https://briankeating.com/blog Audio-only podcast: https://briankeating.com/podcast Learn more about your ad choices. Visit megaphone.fm/adchoices

Transcript

0:00
1:06:16

11,625 words · 5 speakers

Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.

Topics

  • OpenAI
  • Simons Foundation

Page updated · transcript and summary are machine-generated