Ferenc Huszár

We have ≥$10k to support talented 14-18 year olds whose studies were interrupted by war in Ukraine. We especially would like to hear from IMO, EGMO, MEMO, IOI, EGOI, IPhO, IChO contestants. If you're one or know one, here's the form to apply: ferenchuszar1.typeform.com/ukrainefund-eng

Call For Tasks for IOAI 2025: ML olympiad for high schoolers. I know you all have strong opinions on the ML topics future computer scientists should engage with to build mastery. This is a high-impact opportunity to help us by proposing tasks. ioai-official.org/call-for-tasks/

Hexameter poetry is rare in English, because it doesn't fit the natural rhythm of language, but there are nice examples: "This is the forest primeval. The murmuring pines and the hemlocks," - HW Longfellow "Gen Z boss and a mini, gen Z boss and a mini" - random startup ceo

At ICML this week? Check out @annabelle_cs's paper in collaboration with @LesterMackey and colleagues on Low-Rank Thinning! ⏰ Tue 15 Jul 4:30 - 7 p.m. PDT New theory, dataset compression, efficient attention and more: icml.cc/virtual/2025/p…

A new post with intuitions behind continuous-time Markov chains, a building block of diffusion language models, like @InceptionLabs 's Mercury and Gemini Diffusion. Touches on different perspectives on Markov chains, connections to point processes + more. inference.vc/discrete-diffu…

Can Transformers do better than Bayesian inference in compositional tasks and in-context learning? Our workshop paper at the Delta workshop on Monday @iclr_conf (also at #AABI), led by @SzilviaUjvary answers in the affirmative. Check it out at openreview.net/forum?id=YGcCR… #ICLR2025

Congrats @DobrikG for defending your thesis.

I'm not sure who still reads X and is active on the butterfly app, but my handle there is now inference.vc I haven't really opened X for quite a while.

Can language models transcend limitations of training data? We train LMs on a formal grammar, then prompt them OUTSIDE of this grammar. We find that LMs often extrapolate logical rules and apply them OOD. Proof of a useful inductive bias. At NeurIPS: nips.cc/virtual/2024/p…

Resharing this from the butterfly app jobs.cam.ac.uk

We're expanding our collaboration with AWS. This includes a new $4 billion investment from Amazon and establishes AWS as our primary cloud and training partner. anthropic.com/news/anthropic…

We're delighted to welcome Professor Hannah Fry to Cambridge! Mathematician and award-winning science presenter @fryrsquared will join us as Professor of the Public Understanding of Mathematics @FacultyMaths, following an announcement @NewtonInstitute 👉 cam.ac.uk/research/news/…

Finally this reddit question can be answered with a reference.

Woah, huge news again from Chatbot Arena🔥 @GoogleDeepMind’s just released Gemini (Exp 1121) is back stronger (+20 points), tied #1🏅Overall with the latest GPT-4o-1120 in Arena! Ranking gains since Gemini-Exp-1114: - Overall #3 → #1 - Overall (StyleCtrl): #5 -> #2 - Hard…

Quick, somebody make a depressive sarcasm from the wrong side of Berlin wall Starter Pack. I need more 🦋 followers.

this looks cool: secular Bayesian seminar series

If you don't want your macbook pro to start producing AGI nanobot juice, cancel your ChatGPT subscription today.

✨ Excited to announce Outer-PPO! 🚀 (arxiv.org/abs/2411.00666) Outer-PPO transforms PPO updates by separating estimation from application, allowing us to add momentum and tweak learning rates in the outer loop 🤯 w/ @fhuszar @j_foerst @benjamin_ellis3 @EdanToledo 🧵👇 1/9

I just found this diagram in the notepad on my office desk. It looks like my hand writing, but I don't remember making this. It appears to be a 3-dimensional analogical embedding of Cambridge fish restaurants 🤷

Pornhub replacing Coca Cola as the symbol of Western freedoms and way of life.