Seewoo Lee

TIL thread

Atari’s original name was Sygygy

You can do polynomial interpolation with OEIS for integer sequences: oeis.org/search?q=108%2…

Mathematically, Heisenberg's uncertainty principle states that there's no function f(x) where itself and its Fourier transform are both concentrated near the origin (to be precise, product of |xf(x)|_2 and |x\hat{f}(x)|_2 is bounded below by a positive constant (1/n)

While I agree in principle, in practice I think AI raises tough questions about what we even mean by "good science" in the context of mathematics. There's an infinite number of true mathematical statements, many of which we can but do not bother to prove because we consider them "routine" and therefore uninteresting. What counts as "routine" is subjective, but I think it approximately means "doable using (only) well-known existing techniques". If AI becomes consistently stronger than humans at certain mathematical skills (for example, "combining already-existing techniques in a new way"), then certain types of previously non-routine problems will become routine in a higher level sense: doable by the well-known technique of querying an AI chatbot. At that point, are those problems still interesting?

Fully endorse this view. I have some worries about how AI tools will change the profession of mathematics, specifically around training. But the fundamental goal has to be to do good science, and clearly AI tools can help with that.

I keep forgetting that I should not try to prove new theorem when I have 2 weeks to submit my thesis

But I do worry that there’s a gold rush mentality and not everything is being checked carefully.

To clarify one aspect of the chat around Erdos problem 1196: Yes, it was open for around 60 years, in that it was asked 60 years ago. Yes, it was a serious problem that mathematicians had worked on. No, mathematicians had not been working non-stop on it for 60 years.

I want to briefly talk about the use of AI in this paper, which I think is the first paper I've written where LLMs had a substantial effect on the final product, as you can perhaps tell from the acknowledgments:

Choose your favorite 2d vector field, and explain why it is <-y/(x^2+y^2), x/(x^2+y^2)>

About a month left until graduation… it took a long time

Jennifer S. Balakrishnan, Francesca Bianchi, Netan Dogra: p-adic elliptic polylogarithms and cubic Chabauty arxiv.org/abs/2604.20662 arxiv.org/pdf/2604.20662 arxiv.org/html/2604.20662

Writing (thesis | code | paper)

Writing (thesis | code | paper | article)

Wow PHM is amazing movie

Congratulations to Vesselin Dimitrov @Caltech and Yunqing Tang @UCBerkeley on winning the 2026 New Horizons in Mathematics Prize for work in Diophantine geometry, including the proof of the Atkin-Swinnerton-Dyer unbounded denominators conjecture and new irrationality results for special values of Dirichlet L-series. breakthroughprize.org/News/98

Very excited to share: I'm co-organizing the Future of Mathematics Symposium at Stanford, held on May 1-2! We have a remarkable line-up of speakers, with Keynotes from Fields Medalists Terry Tao, Maryna Viazovska, and Michael Freedman. Free pre-registration link below

1/ Mathematicians send us their real, cutting-edge research to "digitalize". PDF to Lean. So we give AxiomProver the task of autoformalizing these papers. Here's our second test case, arising from the study of plane curve singularities in algebraic geometry -- by @Y_F_Huang 💫

Some dismiss Erdős problems as trivialities - this couldn't be further from the truth! While many are amusing novelties, some of them are the most central problems in number theory and combinatorics. A blog post with, in my view, the 10 most important: erdosproblems.com/forum/thread/b…

Worried about Anthropic's Mythos? Fully formally verified code generation is the defense. Combining Lean, frontier models, multi-agent scaffolds, and inference scaling, we show <12mo benchmarks jumping from 20% to 70%. Real-world verification is here. facebookresearch.github.io/wybecoder/ 1/

GPT-5.4 Pro solves Erdős Problem #1196! Very pleased with this result; definitely my favourite thus far! This problem has been thought about for some time which makes this reasonably impressive and meaningful (see Lichtman's comments below). Formalisation is underway!

I hope people also remember that there was Sedol Lee's move 78