Come visit our poster "SwitchHead: Accelerating Transformers with Mixture-of-Experts Attention" on Thursday at 11 am in East Exhibit Hall A-C on #NeurIPS2024. With Piotr Piękos, Kazuki Irie and Jürgen Schmidhuber.

We’ve released updated 2024 lectures from my Stanford RL CS234 class! shorturl.at/pf427 This includes a guest lecture on Direct Preference Optimization (DPO) from 1st authors Rafael Rafailov @ NeurIPS Archit Sharma Eric tinyurl.com/3kr4czth Thanks to Stanford Online!

new stuff for developers today, including o1 in the API with new features like structured outputs and function calling, realtime API updates, preference fine-tuning, and more.

Our paper on predicting the emergence of crystals from amorphous precursors with deep learning potentials is now published in Nature Computational Science! 🎉 Google DeepMind

"Every major AI company says it aspires to bring positive change to countless aspects of life, but Google DeepMind is doing an uncommonly high percentage of the heavy lifting." Fast Company’s Global Tech Editor Harry McCracken recently spent time with our CEO @DemisHassabis and

o3 represents enormous progress in general-domain reasoning with RL — excited that we were able to announce some results today! Here’s a summary of what we shared about o3 in the livestream (1/n)

also: the model we used for all of our o3 evals is fully general; a subset of the arc-agi public training set was a tiny fraction of the broader o3 train distribution, and we didn’t do any additional domain-specific fine-tuning on the final checkpoint

Our Director in Research Gregory Wayne is one of the team behind Project Astra: our research prototype exploring future capabilities of a universal AI assistant. ✨ He joined our podcast host @fryrsquared to share more about how it was developed and what might come next. ↓

what would you like openai to build/fix in 2025?

🤩Forget MoCap -- Let’s generate human interaction motions with *Real-world 3D scenes*!🏃🏞️ Introducing ZeroHSI: Zero-Shot 4D Human-Scene Interaction by Video Generation. No training, No MoCap data! 🧵1/5 Web: awfuact.github.io/zerohsi/

common themes: AGI agents much better 4o upgrade much better memory longer context “grown up mode” deep research feature better sora more personalization (interestingly, many great updates we have coming were mentioned not at all or very little!)

Building Large Language Models from Stanford 1:45 hours long and best for beginners

insane thing: we are currently losing money on openai pro subscriptions! people use it much more than we expected.

Excited to announce the ICLR 2025 Workshop on Human-AI Coevolution with Mykel Kochenderfer, Diyi Yang, Sanmi Koyejo, David Hsu & more! #AI #HumanAIInteraction #ICLR2025 #AIResearch #EthicalAI #HCI

Imagine an AI agent leading a team of AI scientist agents to tackle challenging, open-ended research, with a human researcher providing high-level feedback. According to Stanford HAI faculty affiliate James Zou, that’s where we’re headed. Read more: stanford.io/407bm9r

it's been fun watching the vibes shift on o1 (including pro) as people learn how to use it

I asked Grok what book is in all 3 libraries... And I started listening to it. This is what I learned...

GM Everybody

Kids pressin their kidnappers is unreal

Embracing AI