Pinned Tweet
Born a few years too late to be a real mathematician
Born just in time to pull on the threads of the universe until they unravel before my eyes
Do not grieve my friends
Today we have lost our pride, but we have gained the world
Jack McCarthy retweeted
continuing the SPDE theme, here is some regularity structures in Lean
dynamical Phi43 on the torus
github.com/scottnarmstrong/R…
I'm enjoying reading TheoremA/challenge.lean
But I can't wrap my head around this definition of white noise. Intuitively, how is it related to e.g. Brownian motion? It's not obvious to me that this is even non-vacuous! (luckily Scott gives a proof)
Any analysts care to explain?
continuing the SPDE theme, here is some regularity structures in Lean
dynamical Phi43 on the torus
github.com/scottnarmstrong/R…
Jack McCarthy retweeted
I have but one comment to add: It is getting f*cking dangerous
Holy cr*p! @kirancodes explains how to get formal program semantics that are provably equivalent to the executed assembly🤯🤯🤯
That’s a real dreamer in action.
Just wait 10 years, every CS student will study her refinement theorem
Only at #LeanNYC
Today is your LAST CHANCE to sign up for Meta’s AI4Math Workshop in NYC!!!
Thank you @arnal_charles and Vivien Cabannes for helping organize this awesome event
Hoping to see you all tomorrow!
docs.google.com/forms/d/e/1F…
Holy cr*p! @kirancodes explains how to get formal program semantics that are provably equivalent to the executed assembly🤯🤯🤯
That’s a real dreamer in action.
Just wait 10 years, every CS student will study her refinement theorem
Only at #LeanNYC
We need a fundamentally new system for evaluating academic work.
How do I tell grad schools that I have 0 journal publications but have been cited many times for my open source contributions?
Thanks Meta FAIR for citing my Formal Conjectures PR fixing a Poincare Conjecture misformalization.
Note: This wasn't a mistake by the FC authors, but in fact an issue with Mathlib upstream! I fixed the same misformalization in Mathlib last month
github.com/leanprover-commun…
The Bitter Lesson for harness design and formal math!
Instead of hand-designing how an AI should prove (decompose, plan, assign agents), what if we just give a frontier model good Lean feedback and let it keep working?
We built h1.
TL;DR:
→ SOTA on formal theorem proving benchmarks
→ can solve open conjectures
→ matches/beats hand-designed workflows
→ one long attempt can beat many retries
[1/7]
Jack McCarthy retweeted
The Bitter Lesson for harness design and formal math!
Instead of hand-designing how an AI should prove (decompose, plan, assign agents), what if we just give a frontier model good Lean feedback and let it keep working?
We built h1.
TL;DR:
→ SOTA on formal theorem proving benchmarks
→ can solve open conjectures
→ matches/beats hand-designed workflows
→ one long attempt can beat many retries
[1/7]
Been thinking about this tweet a lot #LeanNYC
If you haven't watched suckerpinch video on Nopert solids you're missing out
youtube.com/watch?v=QH4MviUE…
The proof is by exhaustion: Use interval arithmetic to eliminate every possible orientation
Lean struggles to prove such results (without `native_decide`)--this proof uses rust
After 2 years and 1000+ hours of self-motivated research, I have finally finished my proof that the Rhombicosidodecahedron is Nopert (does not fit through itself), settling one of the main open problems in a 333-year-old topic.
github.com/bence-hervay/nope…
Another banger from @scottnarmstrong
I'm reading through the Lean repo now. They document some differences between the code<->paper.
(e.g. take p<∞ instead of p≤∞)
Question: Why wasn't the AI able to make the formalization match exactly one-to-one? Did it hit a block?
【偏微分方程式】楕円型方程式の正則性を支えるDe Giorgi–Nash–Moser理論を、退化・非有界な係数まで拡張した研究です。ハルナック不等式などを最適な係数範囲で証明しています。10/8公開。査読前です。#数学
arxiv.org/abs/2610.11879
Today I released a preprint answering a 60-year old problem dating back to Bourbaki. I was originally motivated to work on this problem because it is needed to define differential forms in Mathlib.
I will give a talk at NYU tomorrow presenting the proof.
github.com/Deicyde/alternati…
SIGN UP FOR LEAN NYC TOMORROW: forms.gle/ptPaJjuMWL7Vyyry8
cc: @justincasher
MORE INFO: lean.nyc/calendar
To clarify, I still have not repaired one of the misconfigured proofs (but I'm working on it!)
Specifically, there's an issue with `DefocusingNLS.lean` which can not be fixed by merely altering the solution. The challenge file itself will need to be changed to pass comparator.
Jack McCarthy @Jack______M has fixed these, 6 were immediate and two had short fixes. github.com/Deicyde/math
I never thought the proofs were wrong. But if you misconfigure your proofs, others can't trust them with extra work... like what Jack just did for free. (Thanks!)
I have re-run comparator without the `definition_names` hole. The "Alex's Cheat" is supposed to fail (the sanity test), and SpinAngle is maybe supposed to as well.
From what I can tell right now, OccupiedOverlap is a slight difference that will need to be repaired. Updating soon
8 of the Lean formalizations in OpenAI's repo are "broken" (trivially provable with wrong proofs):
blog.ohaithe.re/post/8299743…
I made a fork of OpenAI's `math` to keep track of all the code that needs to be fixed. cc: @Robertljg
github.com/Deicyde/math