Rob Sneiderman

building AI for automated research and scientific discovery: reasoning agents, and machine-assisted mathematics.mcgill math, stats. sf 2027

[Rob Sneiderman @Rob_Sneiderman]

I’ll be at OpenAI DevDay in San Francisco on September 29.

The transition from increasingly capable models to systems that can accelerate scientific discovery and AI research is the technical problem I care most about. Looking forward to learning from the teams working closest to it.

[Rob Sneiderman @Rob_Sneiderman]

@ziv_ravid Not a pure LLM!

[Rob Sneiderman @Rob_Sneiderman]

@PI010101 @the_rajaj Congrats!

[Paata Ivanisvili @PI010101]

Counterexamples to the Landis conjecture in dimensions 3 and higher. https://t.co/gtdw7xHP2Y

[Rob Sneiderman @Rob_Sneiderman]

@krzyzanowskim @michalmalewicz $2000 can make a killer app though

[Fanghui Liu @Fanghui_SgrA]

🚀 Introducing StatsMLlib: Probability, Statistics, and Machine Learning in Lean 4

🌐 https://t.co/Wz9Osse4QE

Join us for contribute!

@jasondeanlee Peter Bartlett, @weijie444 @btreetaiji Yuekai, Aleksandar, Sho, @yuanhezhang6, Kei, Kazumi, Naoto, Yuma, @KevinHanHuang1

[Rob Sneiderman @Rob_Sneiderman]

@Fanghui_SgrA Much appreciated, Fanghui! StatsMLLib’s and your group’s broader results are very interesting and useful, I hope to learn from it.

[Rob Sneiderman @Rob_Sneiderman]

@TheSeaMouse For sure helpful, but those customers will not be valid for actual simulations, and your experiments won’t mean much, this is what @simile_ai is building a solution for it sounds like

[Dhanya Sridhar @dhanya_sridhar]

📣 We are organizing the first InterpScience Workshop @ NeurIPS 2026 in Sydney! The goal of the workshop is to build a more rigorous scientific foundation of LLM interpretability.

📝 Papers due: Aug 28
🌐 https://t.co/lWJOhgZLuI
✉️ [email protected]

[1/7]

[Rob Sneiderman @Rob_Sneiderman]

Tao gives explicit coordinates. Conceptually, set h = va + ub and k = -2vb + uc. Since r = vh + uk, projection from the common point sends H and R to a constant section and the diagonal in P1 x P1.

Their complement is A2, so X is an A1-bundle over A2. Such a bundle is trivial, hence X is A3.

This is essentially the Sawin-Speyer-Litt argument from what I understand from back and forth with GPT

[Rob Sneiderman @Rob_Sneiderman]

@neetcode1 it is going to happen

[OpenAI @OpenAI]

We’re releasing the manuscripts, formal Lean certificates, and reasoning walkthroughs so mathematicians can examine these results and build on their ideas. https://t.co/oDT2J8F6Ez

[Rob Sneiderman @Rob_Sneiderman]

@jynx_29 Very good video actually

[Rob Sneiderman @Rob_Sneiderman]

@KL_Div as soon as I read 'we discovered a third pretraining axis' I knew this would have multiple issues

[jacob tsimerman @Jacob_Tsimerman]

A couple of websites

I want to share a couple websites aimed specifically at mathematicians regarding AI-safety and AI-risk .

The first one I made myself: https://t.co/Ye2ZF5CmJj. This is designed to be a very high level and accessible resource describing some research agendas, collections of problems, and organizations in the field of AI-safety. I hope that for mathematicians considering directing some of their effort in this direction, this eliminates some of the friction. I plan to expand it quite a bit, but I hope it's already useful! I'd be happy and grateful to hear (constructive) feedback on it :-)

1/2

[Rob Sneiderman @Rob_Sneiderman]

Exactly. The hard part is often not learning one more framework; it’s having the math in the right order. There is a rigorous source grounded Math → ML track for that: calculus, linear algebra, probability, learning theory, optimization, then deep learning.
https://t.co/yw3njkPHuv

[Rob Sneiderman @Rob_Sneiderman]

@krishnanrohit What’s happening with BenchBenchBench

[Rob Sneiderman @Rob_Sneiderman]

@__alpoge__ @ElliotGlazer blocked me for pointing out this fact I think. It is still true though. FrontierMath also has major issues.

[Rob Sneiderman @Rob_Sneiderman]

@__alpoge__ Doesn’t seem legit, if a “generic prompt” from fable can solve them, why not post the “generic prompt” ever or the chats/cot?

[Rob Sneiderman @Rob_Sneiderman]

@GaryMarcus Can you explain the PDF point? Multiple models already can extract PDF at superhuman levels

[Jiahui Liu @jiahuiteddy]

I'm actually less shocked about the open problems solved by AI and more shocked about how many people truly believe research is about competing to get the results, instead of learning for the sake of our own curiosity and meaning.