Liu Yao 刘杳

https://mathstodon.xyz/@liuyao

AI ⋉ Math. Blogs at Observable on interactive math.

web: https://observablehq.com/@liuyao12
github: https://github.com/liuyao12/ConvNets-PDE-perspective

Public Key

npub1psrkxewqq78p8nngpcttgqa8lc38hlsf3r6dck4dc3kl43g3cv4qwjgeeh

NIP-05 Address

liuyao_at_mathstodon.xyz@momostr.pink

Profile Code

nprofile1qqsqcpmrvhqq0rsnee5qu945qwnlugnmlcyc3axut2kugm06c5gux2spz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dsy457qt

Publishing to

nostr.now

Author Public Key

npub1psrkxewqq78p8nngpcttgqa8lc38hlsf3r6dck4dc3kl43g3cv4qwjgeeh

Show more details

Last Notes

2026-10-12 02:41:00 UTC

- reply

Interesting. Strange that it's got an acronym as the title of its wikipedia page. I asked AI for pronunciation, and apparently it's "seeks" or "seeks pee-oh-vee-em". Can't we find a better name for it?

2026-09-22 07:23:44 UTC

- reply

I presume you don't have a strong aversion to feeding such questions to an LLM. (No explanation of the same kind, if you don't mind a little spoiler.)
https://chatgpt.com/share/6ab22c08-5c10-83ea-85cd-4bf8da82a093

2026-08-20 06:37:00 UTC

- reply

I wanted to see if AI can build an interactive visualization for this. Still not happy with their 600-cell, but here's 120-cell: https://liuyao12.github.io/3-sphere/?mode=120-cell
https://media.mathstodon.xyz/media_attachments/files/117/126/446/740/484/832/original/fc7bd2194524b1d2.png

2026-07-08 06:45:33 UTC

- reply

seems perfect for AI to work out

2026-06-15 15:09:57 UTC

- reply

Another resource is lectures by Venkatesh https://www.youtube.com/watch?v=qMTUlD6SnIc

2026-06-12 03:32:37 UTC

- reply

Sounds like translation surfaces but with "ridges"?
https://en.wikipedia.org/wiki/Translation_surface

2026-05-22 01:42:15 UTC

- reply

Will Sawin was one of the people who checked and distilled OpenAI's proof. It's a safe bet that he had come to the explicit bound before the release, and chose to publish separately and simultaneously.

2026-04-27 05:40:51 UTC

- reply

Presumbaly this does not cover quasicrystals/aperiodic (mono-)tilings?

2026-04-27 03:40:47 UTC

- reply

Appreciate your efforts! This has very much to do with Conway's (or Thurston's) orbifold notation of space group that I hope you will get to :)

2026-04-25 06:44:20 UTC

- reply

That's so much more clear than other accounts I've seen.
Just to make sure: in L -> G -> P (kind of funny you put 0 on one end and 1 on the other), L is understood as "translations", and P would be those fixing a point, hence "point group", right?

2026-03-14 13:20:06 UTC

- reply

ArXiv already faces real challenges from AI generated papers, and this doesn't seem to be the way to fence off what's coming.
I was just thinking the other day that anything you can prompt AI to generate within a few minutes would not be worth publishing. Existing papers should also undergo an evaluation by AIs (one AI reads it, summarizes the main ideas, and feeds to another AI to reconstruct.) And we should probably only archive the prompts.

2026-02-25 01:58:40 UTC

- reply

Presumably sundials only work at specific lattitude? The Chinese ones can work anywhere, you just have to tilt it differently https://commons.wikimedia.org/wiki/File:Beijing_sundial.jpg

2025-10-31 05:35:21 UTC

- reply

How many books is that payment equivalent to, in terms of money you actually would have received?
I can't speak for the authors, as I have never written any books. But I feel that the AI models trained on these books are still valuable, just that they can't make money out of them. If only there's a way to pay you each time a query (or the AI's response of it) references your books.

2025-10-16 02:08:23 UTC

- reply

Does that means one should avoid travelling through the south Atlantic? Not that it's on my future plans.

2025-09-25 02:30:45 UTC

- reply

Most mathematicians who are drawn to science would naturally favor physics, while chemistry seems like a mess (too many things to memorize, from high school experience). Perhaps crystollography would be an exception. Knowing your interests in entropy, I'd bet the "concept" of entropic bond would be your cup of tea: https://www.youtube.com/watch?v=q9_2UR1DsgA

2025-09-05 12:22:41 UTC

- reply

There you have Bragg the Elder 😀

2025-08-05 22:52:39 UTC

- reply

I vividly had this thought when I was like 10, an infinitely long string of zeros, followed by a 1. I thought we all had to struggle with reconciling ourselves about this fact at some point (not necessarily having to involve definition/construction of reals), and apparently it's worse for grown-ups.
The best argument I heard, though I was mature enough when I heard it, is that "if they are truly different, tell me a number that's in between these two."

2025-06-30 06:34:16 UTC

- reply

they've kept it alive from the days of Pythagoras, Euclid, and Archimedes.

2025-05-26 01:16:07 UTC

- reply

I remember this image on the door of Bill Thurston, in the last years of his life. Sharing w/ @npub1nf4…nqe4

2025-05-21 03:59:31 UTC

- reply

they've been doing great conservation work for a number of years. They have a website https://homingleopards.org/

2025-02-28 07:13:08 UTC

- reply

not the same sort of number (cardinal, not ordinal) https://en.wikipedia.org/wiki/15_and_290_theorems
For theorem labels that appear in a paper then got stuck, I offer Zariski's Main Theorem
https://en.wikipedia.org/wiki/Zariski%27s_main_theorem

2025-01-29 01:41:06 UTC

- reply

I don't know why people are not talking about their prior work on math. Maybe they deliberately didn't mention that in their new reports?
https://media.mathstodon.xyz/media_attachments/files/113/909/092/716/809/865/original/c073e7f31d6f450b.png

2025-01-26 02:04:48 UTC

- reply

Would be interesting to note that they are somewhat known in the AI+proof assistant community, as the top winners of last year's AIMO competition had used their math model as the base model.
I had make contact with someone on their team, and got the impression that the company (with its background in hedge fund) had the vision to work on math related AI in the hope that it would pay off in their general model. I did not expect it to work, not to say so fast.
https://media.mathstodon.xyz/media_attachments/files/113/892/196/809/780/456/original/7c354a7b96c1992a.jpeg

2025-01-08 22:49:05 UTC

- reply

yes, I remember your posting about it on the other site.

2025-01-08 07:52:40 UTC

- reply

Thanks for reminding me to get back on with a little project of mine, which lets you compute digits in your browser. Not there yet to do the AGM algorithm, but it occurs to me that you can design a whole course on calculus just on the various ways to compute pi. (I'm never a fan of the pop math side of it, but apart from the digits, pi is indeed awesome.)
https://mathstodon.xyz/@liuyao/113780063926562388
https://media.mathstodon.xyz/media_attachments/files/113/791/644/771/685/309/original/00728a25dff9a70e.jpeg

2024-12-18 02:11:45 UTC

- reply

if i'm allowed to think out loud: it looks like the three cynlinders are carefully positioned (touching on the outer surface), and it probably has to depend on the width of the Mobius band or you'd be able to widen it indefinitely. A self-intersecting band?

2024-12-16 07:13:51 UTC

- reply

how did it know where to go? the article didn't give a clue.

2024-10-26 09:01:05 UTC

- reply

I tried to use ChatGPT o1-preview to do some coding (for a different problem). Not too bad, but still requires a few back-and-forths. I'm hoping the full version of o1 is better at this kind of stuff. https://observablehq.com/@liuyao12/tiling-with-lattice-polygons

2024-10-09 03:33:52 UTC

- reply

now they have no excuse not giving one to Ed Witten. He took ideas from physics and led to many great advances in mathematics!