Gabriel Poesia
@GabrielPoesia
Followers
821
Following
369
Statuses
131
CS PhD student @Stanford Learning models that do math to help human learning Mineirinho from Brasil
Stanford, CA / Belo Horizonte
Joined November 2015
Now, for LLMs with potential recall of The Whole Web, finding "new" problems is a ton harder. Recalling and adapting the solution of the right one of millions of problems is still an extremely impressive/hard skill. Just maybe not exactly the skill we thought we were evaluating.
0
2
24
Posted. Then I found a made-up citation to a NeurIPS paper (would be cool if it existed). Added a comment pointing it, and questioning if it was all LLM-generated. Withdrawn 1 min later. Waste of time... This still feels rare in my reviews (luckily), but I guess it must happen
0
1
6
If you're at NeurIPS, come tomorrow for the Oral+Poster on "Learning Formal Mathematics from Intrinsic Motivation"! Really fun work with @DavidKarlBroman @nickhaber @noahdgoodman that put together much of what I did in the past years, w/ a new twist with open-ended learning!
Excited that @GabrielPoesia will be presenting his Oral on Learning Formal Mathematics From Intrinsic Motivation. We make and prove conjectures from scratch, without any human data, by learning what is hard but provable. Gabe’s on the job market, btw.
1
18
80
RT @oh_that_hat: The workshop on mathematical reasoning and AI is back! - compare human-level mathematical reasoning to current AI - evalu…
0
40
0
RT @mdoumbouya: We introduce h4rm3l, a language, a synthesizer, and a scalable and explainable dynamic LLM red-teaming toolkit. h4rm3l foun…
0
8
0
RT @minimario1729: 📢Excited to announce the 4th MATH-AI Workshop at #NeurIPS2024 to discuss some of the exciting recent advances in AI for…
0
12
0
@nileshtrivedi @zli11010 Hey Nilesh. Thanks! I just replied on Github with updates. I'm working on the code release for the mentioned paper (, where I also extended Peano with backward actions, and focus more on proofs. Will tweet about it in the coming week once the code is out!
1
0
1
RT @ericzelikman: Will present Hypothesis Search @ICLR this morning on LM inductive reasoning (specific examples→general rules). We show co…
0
20
0
RT @megha_byte: Finally used @midjourney to finish an idea I've had for a while: Re-creating scenes from some of the wacky dreams ✨ I had d…
0
2
0
RT @rose_e_wang: What do great teachers do to be good at teaching? What can this teach us about LMs? Most of us experience the “front stag…
0
19
0
RT @ericzelikman: Did you know there’s a task people easily solve but GPT-4 fails? From a few input-output grids, ARC asks you to infer and…
0
70
0
@FWS_Dev @OrenElbaum Yes... This is not the first instance of this happening with this guy. But we did release our code on @OrenElbaum there's a simple demo guide to illustrate & get started, plus LogicGuide. Let me know if anything is unclear / setup breaks or is confusing!
0
0
0