korrents

korrents · video

The last IMO problem AI could not solve

Grant Sanderson · 51m · youtube.com ↗

10 korrents from this recording

Grant Sanderson did not write this page.

Every claim below is a statement made in this recording, quoted word for word and linked to the second it was said, so you can hear it rather than take our word for it. The wording comes from the transcript published alongside the recording; the sentence above each quote is our reading of the claim, not their wording.

  1. 2:37 · watch on youtube.com↗

    The IMO almost feels like ancient history as a benchmark as everything has speedrun its way into the domain of research math over the last year.
  2. 1 min later
  3. 3:42 · watch on youtube.com↗

    Models, at least as I've interacted with them, don't really seem to have a great sense for what makes a problem or a strategy beautiful. and finding the solution here absolutely benefits from an appreciation of beauty.
  4. 3 min later
  5. 6:55 · watch on youtube.com↗

    Now, if you step back, I think a very important general lesson about problem solving is that what can look like genius is usually the residue of experience. Insights don't come from nowhere.
  6. 16 min later
  7. 22:44 · watch on youtube.com↗

    I often say that math rewards you when you respect its symmetries. And what we're seeing right now is how it punishes you if you don't.
  8. 24 min later
  9. 46:30 · watch on youtube.com↗

    we spent a lot of time with very visual intuition. I imagine that's harder to train for than purely symbolic reasoning.
  10. 2 min later
  11. 48:29 · watch on youtube.com↗

    Now, finding that kind of narrative to me is deeply rewarding. It's very creative. I always feel like I come out understanding a piece of math much better than I did simply by understanding the proof.
  12. 1 min later
  13. 49:05 · watch on youtube.com↗

    And the role of proofs and open problems has always been one of a guiding light to help people know where and how to push forward that understanding.
  14. 1 min later
  15. 49:35 · watch on youtube.com↗

    But what's useful is not the proof or even the result. It's all of that surrounding understanding and the spillover effects that it can have into the engineering world. So if you're able to generate proofs that don't necessarily come with human understanding, that kind of completely defeats the point.
  16. 49:54 · watch on youtube.com↗

    Because the idea of using proof as a proxy [music] is so deeply ingrained into what we give academic credit for that it exposes a very stark need for a new distinct way to measure whether someone's published output really does push forward human understanding.
  17. 50:16 · watch on youtube.com↗

    I think that we as a community should try to formalize what we mean by a motivated explanation and then we should elevate it to have the same status and give the same kind of academic credit that proof historically has.