A Waymo vehicle was driving in a 25mph zone in LA when an oncoming car swerved into its lane while speeding up to over 70mph… 3x the speed means 9x the destructive energy. Waymo vehicle reacted safely. Source: Dmitri Dolgov (Co-CEO at waymo)

Awesome LLM Post-training This repository is a curated collection of the most influential papers, code implementations, benchmarks, and resources related to Large Language Models (LLMs) Post-Training Methodologies. github.com/mbzuai-oryx/...

Thanks for reposting my video. I think Waymo also solves the critical urban problem of literally not having people dying every day for no reason

How to Scale Your Model This book aims to demystify the science of scaling language models on TPUs: how TPUs work and how they communicate with each other, how LLMs run on real hardware, and how to parallelize your models during training and inference so they run efficiently at massive scale.

That immigrants, from China, India, Iran, Latin America, and so many places choose to come here is a blessing and a gift. That this needs to be said and that politicians ever make them feel otherwise is an eternal disappointment.

A vision researcher’s guide to some RL stuff: PPO & GRPO by Yuge (Jimmy) Shi This is a deep dive into Proximal Policy Optimization (PPO), which is one of the most popular algorithm used in RLHF for LLMs, as well as Group Relative Policy Optimization (GRPO) proposed by the DeepSeek folks.

The PyTorch developer's guide to JAX fundamentals cloud.google.com/blog/product...

#2025 is the sum of the first 9 cubes 🤩 1+8+27+64+125+216+343+512+729 🥂 🎉 🎆 🎇 🧮

Would be fascinated to learn how this paper came into being given the authors on it: arxiv.org/abs/2412.05747

Applications open on Dec 20 for the #Research Scholar program, which aims to strengthen long-term collaboration with the academic community by supporting early-career professors pursuing research in fields relevant to #Google. Learn more & apply by Jan 27 ↓ research.google/programs-and...

A good advice from Victor Dibia

A short list of tips for keeping a clean, organized ML codebase for new researchers: eugenevinitsky.com/posts/quick-...

If machine learning is the high-interest credit card of technical debt, quantization is the back alley predatory loan

Gemini 2.0 is out, and there's a ton of interesting stuff about it. From my testing it looks like Gemini 2.0 Flash may be the best currently available multi-modal model - I upgraded my LLM plugin to support that here: github.com/simonw/llm-g... Gemini 2.0 announcement: blog.google/technology/g...

Going full circle: in 2019 Waymo made an April Fools video about a dog-only robotaxi service. Now people are actually sending their dogs in Waymo unattended. youtu.be/ljbeFpOHvEA?...

Reinforcement Learning: An Overview This manuscript gives a big-picture, up-to-date overview of the field of (deep) reinforcement learning and sequential decision making, covering value-based RL, policy-gradient methods, model-based methods, and various other topics. arxiv.org/abs/2412.05265

Happy Birthday to Gemini! ✨🎂 Around this time last year, I was working on the Gemini Launch and it was exciting to have access to such models After one year I've learned a lot and I'm still amazed of what can be done! best feature: 2M context window 🤯 developers.googleblog.com/en/looking-b...

I’m late to the game here, but super impressed this UI is an open source React Native app, the framework has come a long way!

Derek Sivers once said “Mastery is the best goal because the rich can’t buy it, the impatient can’t rush it, the privileged can’t inherit it, and nobody can steal it. You can only earn it through hard work. Mastery is the ultimate status.” Does it still hold in the age of LLMs? 😎

Paligemma2 is out! Bigger models, better results. For the best experience, do not forget to finetune. Congrats Paligemma2 team!

Hi Bluesky! Any AI/ML following recommendations?

Is this a correct description of how fields cross-pollinated each other? - CV and speech jump-started deep learning - NLP innovated with transformers, which were then copied by CV and others - Robotics and CV embraced RL early on - NLP now discovers RL, first through RLHF and now for planning.