Can LLMs do reinforcement learning in-context - and if so, how do they do it? Using Sparse Autoencoders, we find that Llama 3 relies on representations resembling TD errors, Q-values and even the SR to learn in three RL tasks in-context! Co-lead with the inimitable Can Demircan

What will a world look like with 100 billion digital human beings? Today we share our tech report on Project Sid – a glimpse at the first AI agent civilization (powered by our new PIANO architecture). github.com/altera-al/proj… 1/8

Currently Deep RL is going through an imagenet moment and very few people are aware. This has major implications for RL applications and anyone interested in modeling behaviour (e.g. Econ and neuroscience). To find out more watch my recent talk @ICML2024: slideslive.com/39022179

A Captivating Close-Up of Saturn's Rings

Hmm I would expect astrology to work at least a little bit since many phenomena in the world are periodic and the astrological charts are basically periodic basis functions... unless all the astrologers are trained on uninformative labels.

Oh I see you're not "addicted to dopamine"

good stuff

This cannot fly😡

#winning_the_software_lottery I'm buying nvidia stocks first thing tomorrow.

Bro is cookin

In this video the /ch/ in German word "China" is clearly pronounced [ç] (or a bit like [x] in the male's voice), while almost all the Germans I've met in real life say [ʃ] or [k], basically "Schina" or "Kina"... has time changed so much?

One of the reason I prefer Bluesky to X (as well as Threads/Facebook/LinkedIN) is simply because my newsfeed isn't controlled by an engagement-driven algorithm This reduces the incentives for gaming the algorithm with clickbait, conflict entreprenuers, etc.

Nice paper! I have actually wanted to do this for years... science.org/doi/10.1126/sc…

Kenneth Stanley Yes, RL can drive open-ended novelty search. It’s finally time to make self-driven, sustained discovery a reality

Neuroscientists' wet dream?

Hey Europe, you want a vibrant tech industry, right? Do this immediately (it's genius!) linkedin.com/posts/yann-lec…

Agents training in RL gym

Wassily Kandinsky's Dance Curves, 1926