We wrote a little #NeuroAI piece about in-context learning & neural dynamics vs. continual learning & plasticity, both mechanisms to flexibly adapt to changing environments:
arxiv.org/abs/2507.02103
We relate this to non-stationary rule learning tasks with rapid performance jumps.
Feedback welcome!
(1/7) New preprint from Rajan lab! 🧠🤖
@ryanpaulbadman1.bsky.social & Riley Simmons-Edler show–through cog sci, neuro & ethology–how an AI agent with fewer ‘neurons’ than an insect can forage, find safety & dodge predators in a virtual world. Here's what we built
Preprint: arxiv.org/pdf/2506.06981
Humans and animals can rapidly learn in new environments. What computations support this? We study the mechanisms of in-context reinforcement learning in transformers, and propose how episodic memory can support rapid learning. Work w/ @kanakarajanphd.bsky.social : arxiv.org/abs/2506.19686
Pleased to share our ICML Spotlight with @eberleoliver.bsky.social, Thomas McGee, Hamza Giaffar, @taylorwwebb.bsky.social.
Position: We Need An Algorithmic Understanding of Generative AI
What algorithms do LLMs actually learn and use to solve problems?🧵1/n
openreview.net/forum?id=eax...
Very proud of @rtpramod.bsky.social and the rest of our team for this lovely work showing that the brain's Physics Network represents object-to-object contact and predicted future events:
Our work, out at Cell, shows that the brain’s dopamine signals teach each individual a unique learning trajectory. Collaborative experiment-theory effort, led by Sam Liebana in the lab. The first experiment my lab started just shy of 6y ago & v excited to see it out: www.cell.com/cell/fulltex...
For almost a decade, there's been a lot of (justified) hand-wringing and paper-writing about fairness issues in AI. This case gets to the heart of a very important question - how much of that work has materially improved the lives of real people?
Grateful for this careful & honest investigation.
Our new preprint from Rajan lab (Harvard):
"Deep RL Needs Deep Behavior Analysis: Exploring Implicit Planning by Model-Free Agents in Open-Ended Environments"
Sophisticated & sometimes insect-like planning, exploration, predator evasion, and foraging strategies by DRL.
arxiv.org/abs/2506.06981
A big challenge for comp social neuro is to go to more naturalistic small groups while still doing controlled, goal-oriented experiment & analysis. We present a strong effort in that direction (from RIKEN) showing how humans balance memory, reciprocity, value. w/ fMRI www.biorxiv.org/content/10.1...
"We find that all five studied off-the-shelf [military-related] LLMs show forms of escalation and difficult-to-predict escalation patterns.. models tend to develop arms-race dynamics, leading to greater conflict, and in rare cases, even to the deployment of nuclear weapons." arxiv.org/abs/2401.03408
"In contrast to past systematic replication efforts.. replication attempts here produced the expected effects with significance testing (P < 0.05) in 86% of attempts... justifies confidence in rigour-enhancing methods to increase the replicability of new discoveries"
www.nature.com/articles/s41...
"We show that even in a simple, idealised network model, many mechanistically different plasticity rules are equally compatible with empirical data... Our results suggest the need for a shift in the study of plasticity rules"
www.biorxiv.org/content/10.1...
Next we created a nonlinear latent variable model of OFC activity in our task using CEBRA, the awesome new method from @trackingactions.bsky.social' lab, to understand how the task is encoded in the neural circuit at the level of aggregate neural dynamics
