New (2h13m 😅) lecture: "Let's build the GPT Tokenizer" Tokenizers are a completely separate stage of the LLM pipeline: they have their own training set, training algorithm (Byte Pair Encoding), and after training implement two functions: encode() from strings to tokens, and

Introducing Gemma: a family of lightweight, state-of-the-art open models for developers and researchers to build with AI. 🌐 We’re also releasing tools to support innovation and collaboration - as well as to guide responsible use. Get started now. → dpmd.ai/3UJu1Y1

Our Proceedings are out! 🎉 Thanks, Javier Burroni and Neil Lawrence🙏! Find them here 👉proceedings.mlr.press/v238/. (Any fixes needed can be submitted by pull request.) Impressed by our papers? Join us in Valencia for more! Stephan Mandt @ AISTATS’25 yingzhen Gavin Kerrigan Jiaxin Shi

I will be presenting our work "Enhancing In-context Learning via Linear Probe Calibration" this week at AISTATS in Valencia, Spain. If you are interested visit us at our poster on Thursday May 02 at 5pm or checkout our paper: arxiv.org/abs/2401.12406

Excellent opportunity for outstanding researchers!

🤞

Silver medal at the IMO 🤯 deepmind.google/discover/blog/…

If you are looking to conduct research full-time on the foundations of AI, and • you have read the RL textbook and done the exercises, • you agree with the Alberta Plan for AI Research, • you already have a PhD, • you are open to spending some time in Edmonton, then the

My GAN co-author Sherjil Ozair has written about some memories of 2012-2014 in the context of GANs winning one of this year's test of time awards, worth a read for the nostalgia if you were around back then, or for learning what it was like if you weren't

Viewing LLMs as systems with latent "skills" and tasks/benchmarks as having "required skills" is a fruitful research perspective inspired by Item Response Theory. The resulting statistical models are interpretable and easy to fit using publicly available LLM evaluation data.

3.5hrs of pure joy! 🙌 Thank you Andrej Karpathy ! 👏

Nice work! 👏

Great job by Soheil Feizi and the team! 👏

Just finished reading the paper—great read! Amazing work Andreas Auer Günter Klambauer Sepp Hochreiter! 👏

Very happy to share that our work "Out-of-Distribution Detection using Synthetic Data Generation" has been accepted at COLM 2025! 🎉 Grateful to have worked with an incredible team Muneeza Azmat, Rå¥å, Mikhail Yurochkin 👏 Conference on Language Modeling #COLM2025