New preprint with Michael C. Frank and Maya Mathur: Eleven years of student replication projects provide evidence on the correlates of replicability in psychology (psyarxiv.com/dpyn6/) reporting 176 replications from students in a grad methods class Stanford Psychology. (1/5)

Excited to share LAnguage Reward Modulation for Pretraining Reinforcement Learning! LAMP💡pretrains a language-conditioned agent without human supervision using VLM rewards and unsupervised reinforcement learning w/ Amber Xie Carlo Sferrazza Younggyo Seo Stephen James Pieter Abbeel

✨ Now out in Psychological Review! ✨ We present a new account of relevance that weighs both **epistemic** and **decision-theoretic** utility: statements are relevant if they improve the listener’s future decision-making. tedsumers.info/_files/ugd/290…

Four years in the making, my main PhD work is in Nature Human Behaviour! Decades of work has shown both benefits and costs to group decision-making. Can we restructure social networking algorithms so there are fewer costs and more benefits? Paper: rdcu.be/dquI1 1/10

I am thrilled to finally share these results with the world! We found that chimps and bonobos likely remember familiar conspecifics whom they haven't seen in years -- possibly as long as 26 years. It's the longest memory ever recorded in nonhuman animals. nytimes.com/2023/12/18/sci…

Our paper on optimal nudging is out in Psych Review! We use resource-rational analysis to formalize nudges and predict their effects. We also show how to use this approach to automatically construct *optimal* nudges that best improve choice. With Fred Callaway and Griffiths Computational Cognitive Science Lab

1/ Today in Science, we train a neural net from scratch through the eyes and ears of one child. The model learns to map words to visual referents, showing how grounded language learning from just one child's perspective is possible with today's AI tools. science.org/doi/10.1126/sc…

Check out our new paper in NatureSustainability led by Rachit Dubey w/ matt hardy and Griffiths Computational Cognitive Science Lab! 🤖💻AI-generated visuals of car-free US cities help improve support for sustainable policies🚵🌳 rdcu.be/dzqjm + v nice News & Views piece: nature.com/articles/s4189…

Language models today are trained to reason either 1) generally, imitating online reasoning data or 2) narrowly, self-teaching on their own solutions to specific tasks Can LMs teach themselves to reason generally?🌟Introducing Quiet-STaR, self-teaching via internal monologue!🧵

I know I am late in the project cycle for this, but I do have suggested edits for the team behind nature.com/articles/d4158… My overall comment is that the central claims in the original are lacking in empirical support.

New preprint! Super excited to share this project with Anne Collins that’s been 3 years in the making. We introduce two algorithmic architectures of how humans learn and represent hierarchically structured decision policies via abstractions and compression. osf.io/preprints/psya…

1/ 🚀 Presenting AGI - Artificial Generational Intelligence 🚀 We apply the concept of cultural accumulation to RL and find that agents can improve across generations, outperforming those trained for a single lifetime of the same experience budget! Co-led w/ Chris Lu. 🧵

Should AI be aligned with human preferences, rewards, or utility functions? Excited to finally share a preprint that Micah Carroll Matija Hal Ashton & I have worked on for almost 2 years, arguing that AI alignment has to move beyond the preference-reward-utility nexus!

I am really beyond words to be able to share with the world what Michiel Bakker, summerfieldlab @summerfieldlab.bsky.social, and I and a truly world-class team Google DeepMind worked on for the past 2 years. Out today in Science Magazine, “AI can help humans find common ground in democratic deliberation”