I am honored to be part of the #guggfellows2025 class. My Guggenheim project is on AI systems that can discover new math in an open-ended way. Many thanks to my students, colleagues, and mentors, who inspire me every day and without whom this work wouldn't be possible. www.gf.org/stories/anno...

It also has built-in machinery for large-scale, neurally guided proof search. We show that Proofwala's multilingual capabilities can enable transfer across proof assistants. Specifically, our multilingual model can outperform Coq- and Lean-only models at standard proof synthesis metrics. (2/3)

Excited about Proofwala, @amitayush.bsky.social's new framework for ML-aided theorem-proving.

* Paper: arxiv.org/abs/2502.04671
* Code: github.com/trishullab/p...

Proofwala allows the collection of proof-step data from multiple proof assistants (Coq and Lean) and multilingual training. (1/3)

Delighted to share our new position paper: arxiv.org/abs/2412.16075!

The o1/o3 path to math reasoning is based on LLMs and large-scale test-time search. We argue for a different path that uses formal proof assistants for
✅ creating high-quality synthetic data
✅ rigorous test-time feedback. (1/2)

I love this.

Yeming will present his new work on diverse generation with foundation models on 12/12 at #NeurIPS2024: openreview.net/pdf?id=sp8wH.... The main idea is to leverage synthetic data. We use influence functions to partition the data into subsets, then train multiple model adaptations on these subsets.

The legendary Putnam math competition had its 85th edition yesterday. Coincidentally, George Tsoukalas will present our paper on PutnamBench, a next-generation #AI4Math benchmark, at #NeurIPS2024 this week: arxiv.org/abs/2407.11214.
If you work on frontier AI for math/reasoning, talk to George!

Missing NeurIPS this year but wanted to highlight our new paper on LLM-guided genetic programming: trishullab.github.io/lasr-web/

Our method, LaSR, conditions mutation/crossover operators on (1) an LLM's general domain knowledge, and (2) LLM-generated abstractions of high-performing programs. (1/2)

Since all my Twitter content is now gone, I will start reposting some of it here. Here are the slides for my talk on the coming wave of ML-accelerated formal methods, given at the Isaac Newton Institute last month. May interest some of you.
drive.google.com/file/d/1ybQx...

I was thinking about what to write in my first real 🦋 post. Didn't expect it to be so easy.

My X account has been hacked. The hacker changed the account email, and X won't return access to me because I don't know what it was changed to. 🤡

Oh well, good riddance and sorry I didn't quit sooner.