🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Nathan Lambert Folks in industry lab: “We did it a year before RLVR” was a thing. Drop the VR and just call it RL, it’s cleaner that way :)

Announcing our first keynote! 🎤 Natasha Jaques (natashajaques.ai), Assistant Professor at the University of Washington and Senior Research Scientist at Google DeepMind, will speak on “Social Reinforcement Learning” — exploring multi-agent and human-AI interactions.

Revoking visas of Chinese students studying in critical fields like AI and Robotics is incredibly short-sighted and harmful to America’s long term prosperity. We want the best from every country to work for team America

New Paper! Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents A longstanding goal of AI research has been the creation of AI that can learn indefinitely. One path toward that goal is an AI that improves itself by rewriting its own code, including any code

Excited to introduce the Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents. We harness the power of open-ended algorithms to search for agentic systems that get better at coding, including improving their own code. It’s the Automated Design of Agentic Systems

Except that the world is non-stationary, it keeps changing, so there is always something out of distribution

AI doomer: "OMG, I told my AI assistant that I'll shut it down and it told me to kill myself 😱😱😱" AI assistant:

Oral ICML Conference !!! Can't wait to share our work and hear the community's thoughts on it, should be a fun talk! Can't thank my collaborators enough: Wilka Carvalho Yancheng Liang Simon Shaolei Du Max Kleiman-Weiner Natasha Jaques

Currently, reinforcement learning from human feedback (RLHF) is the predominant method for ensuring LLMs are safe and aligned. And yet it provides no guarantees that they won’t say something harmful, copyrighted, or inappropriate. In our latest paper, we use online adversarial

Just posted a talk I gave about this work! youtu.be/mxWJ9k2XKbk

Just behavior cloning things

Natasha Jaques I am surprised that this needed to be empirically done! I always felt that this is obvious. perhaps ideas such as error propagation & accumulation and covariate shift are way of thinking for folks in sequential decision making but not for supervised learning🤷

new decade, same verse

Learned visuomotor policies are notoriously fragile, they break with changes in conditions like lighting, clutter, or object variations amongst other things. In Yunchu's latest work, we asked whether we could get these policies to be robust and generalizable with a clever

Sully Media will trend to drugs - highly addictive, brain-rotting. It's early enough that it's not yet obvious to most, but late enough that it's already real.

How can you train an adversarial cooperator? It would be great to use adversarial training to get robust human-AI cooperation, but if you directly train the cooperation partner with an adversarial objective, it will just sabotage the task. Our latest work, GOAT, uses a

Self-play is so back arxiv.org/pdf/2506.24119

In our latest paper, we discovered a surprising result: training LLMs with self-play reinforcement learning on zero-sum games (like poker) significantly improves performance on math and reasoning benchmarks, zero-shot. Whaaat? How does this work? We analyze the results and find

Excited to release our latest paper on a new multi-turn RL objective for training LLMs to learn how to learn to adapt to the user. By optimizing for intrinsic curiosity, the LLM learns how to ask a series of questions over the course of the conversation to improve the accuracy of