New on @berkeley_ai blog by @seohong_park: divide-and-conquer value learning for off-policy RL. No TD bootstrapping. Scales to long horizons. https://t.co/VLspSfmRmW 🐻📄

🚨 We wrote a new AI textbook "Learning Deep Representations of Data Distributions"! TL;DR: We develop principles for representation learning in large scale deep neural networks, show that they underpin existing methods, and build new principled methods.

🚀 Excited to release the full inference code of 🧢CAP4D🧢! Generate animatable 4D avatars from any image(s) + driving video. 🤩Also works on stylized photos! 👉 Code: https://t.co/DFJmZHKCtB 🏠 Project page: https://t.co/l6hRa5jYko

Humanoid motion tracking performance is greatly determined by retargeting quality! Introducing 𝗢𝗺𝗻𝗶𝗥𝗲𝘁𝗮𝗿𝗴𝗲𝘁🎯, generating high-quality interaction-preserving data from human motions for learning complex humanoid skills with 𝗺𝗶𝗻𝗶𝗺𝗮𝗹 RL: - 5 rewards, - 4 DR

Genuinely impressive results, fairly traumatizing introduction.

❓How can humanoids learn to squat and open a drawer? Reward-tuning for every such whole-body task is infeasible. 🚀Meet DreamControl: robots "dream" how people move and manipulate objects in varied scenarios, practice using them in simulation, and then act naturally in the

🏓🤖 Our humanoid robot can now rally over 100 consecutive shots against a human in real table tennis — fully autonomous, sub-second reaction, human-like strikes.

REALLY great talk by @chelseabfinn at @ycombinator AI SUS on @physical_int and the future of hardware agnostic generalist models for robotics

Honored to be invited to #AIStartupSchool by @ycombinator in SF this week! The speaker lineup is unreal: Elon Musk, Sam Altman, Garry Tan, Satya Nadella, François Chollet, Andrej Karpathy, Andrew Ng, Fei-Fei Li, John Jumper, Chelsea Finn, Jared Kaplan & more. If you’re in town

At @theworldlabs, we built a new Gaussian splatting web renderer with all the bells and whistles we needed to make splats a first-class citizen of the incredible @threejs ecosystem. Today, we're open sourcing Forge under the MIT license.

1/6🤖How do robots beat professional racers? Racing isn’t just about speed—it’s strategy! 🚨Introducing 🏁α-RACER🏎️: a real-time algo to compute Nash equilibrium strategy, enabling optimal decisions against competitive opponents ✅Scalable, fast, and game-theoretically sound

our new system trains humanoid robots using data from cell phone videos, enabling skills such as climbing stairs and sitting on chairs in a single policy (w/ @redstone_hong @junyi42 @davidrmcall)

I'm at ICLR this week! I'll be presenting ToST, a (provably) computationally efficient high-performance deep architecture derived from information theory and convex analysis principles. 📅 Saturday April 26, 10AM-12:30PM 📌 Hall 3 + Hall 2B #145 💡Awarded a Spotlight! (1/3)

1/6❓Can robots learn to play competitive sports by observing expert human players? 🚨Introducing LATTE-MV, a scalable 3D reconstruction system that⚙️800+ hrs of Youtube videos to create the largest 3D 🏓TT dataset! ✅27 hrs of gameplay | 73k exchanges 🌐 https://t.co/55uNyHu6yE

Vision-language models can control robots, but what if the prompt is too complex for the robot to follow directly? We developed a way to get robots to “think through” complex instructions, feedback, and interjections. We call it the Hierarchical Interactive Robot (Hi Robot).

Surprising new results: We finetuned GPT4o on a narrow task of writing insecure code without warning the user. This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis. This is *emergent misalignment* & we cannot fully explain it 🧵

People have been wondering which model will be the first to reach a 1400 Elo score; no one believed it would be Grok one year ago, but @xAI made it happen!

our latest work on sim2real for dexterous robotic tasks is out. we train RGB-only policies from scratch in sim and have them generalise to real. check Ritvik’s main thread for details

max layers, max cheer 🎄 #merrychristmas community

Hi #NeurIPS2024! 1- Explore ~4,500 NeurIPS papers in this interactive visualization: https://t.co/LjsewLqDye (Click on a point to see the paper on the website) Uses @cohere models and @leland_mcinnes's datamapplot/umap to help make sense of the overwhelming scale of NeurIPS.