PG&E vs Silicon Valley Power > Rio 3.5 Open 397B is a frontier-class general-purpose AI model developed by IplanRIO, the municipal IT company of Rio de Janeiro's city government. huggingface.co/prefeitura-ri…

This is probably the least informative example of monospace fonts you could have come up with. en.wikipedia.org/wiki/Monosp…

I will be covering this and much more in my PyCon Singapore talk! Come say hi! 👋 pycon.sg/

Pop a couple padding tokens in the vocab and call it Fable 5.1

Blog post about my recent optimal tokenizer exploration blog.aqnichol.com/2026/06/10…

timeline feels like a l'oreal commercial

We'll be hosting a tutorial at EAMT (already next week!), KONVENS and MT Marathon on human evaluation. Come learn with us! With support of @maikezufle and @PSchmidtova

One funny thing about ChatGPT is that when I ask it some technical question and it starts doing the "searching the web" and then flashing through websites, it always does some weird ones like imbd. Idk what you are expecting to find there brother but you do you.

They followed up.

Lore.. @marco_computers

📢 CfP for the 2nd version of MOSS at @COLM_conf! sites.google.com/view/moss-c… (Deadline: 6/30) We welcome submissions on small-scale research for algorithmic innovation and scientific understanding across training, architecture, data, evaluation, interpretability, safety, and more!

Requesting that everyone call me Dr. Parameter-Efficient Korean Character-Level Language Modeling for now on. I'm honestly so disappointed in spammers these days. No excuse in the age of LLMs.

Workshop on Responsibly Enabling Data for Foundation Models at #COLM2026 October 9 in SF "Unlocking sensitive data sources responsibly for the next generation of AI" - Amazing invited speakers 😍 - Submission deadline: June 23 🗓️ - Do *you* want to be a PC member? 🫵 @COLM_conf

I am looking to hire 1-2 interns to work with me in Tokyo at RIKEN, starting October 2026. Topics are in intersection of optimization, DL and Bayesian methods. The internships are fully funded. More details here, please share with interested students: moellenh.github.io/variation…

What if you could take three completely different model families… and distill them into one tiny model? 🤯 📜 Paper: arxiv.org/pdf/2605.21699 MOPD (Multi-Teacher On-Policy Distillation) has become a standard procedure in post-training. We already distill multiple specialized variants of the same model into a single set of weights. But what if we could go further - and distill models from entirely different families? Turns out, it is possible. Today we’re releasing a paper on cross-tokenizer distillation - our first steps in this exciting direction. 📄 We distilled Qwen3-4B, Phi-4-Mini, and Llama-3B into Llama-3.2-1B. MMLU jumped from 32.05 → 46.32 when using multiple teachers. 📈 The team is now working on Nemo-RL integration so the community can try this method in their own settings. Plus, we are scaling experiments up. 🚀

Required reading for anyone doing any work with agents

Happy to share that the unprocessed results and code for fitting scaling laws and plotting are now available at: github.com/facebookresearch/…

Maia-3 is here! Our latest models predict what players from 600–2600 would actually play (not just what engines think is best) New SOTA accuracy, more efficient, free, open source. Models: huggingface.co/collections/U… Code: github.com/CSSLab/maia3 Blog: lichess.org/@/ashtonanderson…

The Claude Code use case that has saved me the most time so far is just "get this slightly out of date ML repo working on my workstation". I just clone the repo, tell Claude I want it slightly modernized with a simple script to do smoke tests, etc. It figures out the dependencies, fixes the inevitable CUDA issues, etc. It legitimately saves me dozens of hours of debugging a month and projects that I would otherwise just not even attempt become immediately viable.

The Maia3 source code was released today! github.com/CSSLab/maia3 arxiv.org/abs/2605.19091