This work has been accepted to #COLM2025. If you are in Montreal this week for COLM and would like to chat about this (or anything related to discovery / exploration / RL), drop me a note!
Poster session 2: Tuesday Oct 7, 4:30-6:30pm
Poster number 68
05.10.2025 19:40 — 👍 1 🔁 1 💬 0 📌 0
Thank you for the shoutout Alison! We actually just arXiv-ed the paper. Attaching the thread below :)
bsky.app/profile/agx-...
16.05.2025 16:48 — 👍 3 🔁 1 💬 0 📌 0
Agent samples (without replacement) from the LM prior at inference time to construct a new prior with higher entropy. It then iteratively prompts the LM to take actions that maximize information gain under the new distribution, and eliminate hypotheses inconsistent with new observations.
Hypothesis sampling allows the agent to correct its “disjunctive bias”, and perform equally well on both disjunctive and conjunctive environments when sufficiently many hypotheses are sampled.
How can we help LMs think more rigorously, like scientists?
We fix this “biased prior” by explicitly sampling a higher-entropy hypothesis distribution, then prompting the LM to maximize info gain under the new distribution. This significantly improves exploration and inference performance!
16.05.2025 16:45 — 👍 1 🔁 0 💬 1 📌 0
(Left) When presented with conjunctive evidence, most LMs tend to prefer disjunctive inferences, similar to human adults. (Right) LM exploration behaviour appears more affected by the underlying causal rule, while children's behaviour do not appear to show significant differences.
Why do LMs have this “cognitive bias”? We compare the LMs’ behaviour to human data, and find that most LMs behave like adults, and less like children who are more receptive to alternative hypotheses. This may suggest that LM trained on adult generated data inherits the same human irrationalities.
16.05.2025 16:45 — 👍 1 🔁 0 💬 1 📌 0
Causal exploration efficiency of different LMs, measured by number of hypotheses remaining after each step in the environment. Lower y axis means agent generated observations that eliminate more hypotheses. The goal is to eliminate all but one hypothesis (which is the true causal relationship).
We evaluate LMs on the classic “Blicket Test” pioneered by @alisongopnik.bsky.social . The goal: assess their abilities to discover and infer causal relationships.
Across a range of models, LMs consistently struggle more with the “conjunctive” (AND) rule, but not the “disjunctive” (OR) rule.
16.05.2025 16:45 — 👍 1 🔁 0 💬 1 📌 0
Example of the Blicket Test experiment. A subset of objects activate the machine following an unobserved rule ("disjunctive" / "conjunctive"). The agent needs to interact with the environment by placing objects on/off the machine to figure out the rule.
Language model (LM) agents are all the rage now—but they may exhibit cognitive biases when inferring causal relationships!
We evaluate LMs on a cognitive task to find:
- LMs struggle with certain simple causal relationships
- They show biases similar to human adults (but not children)
🧵⬇️
16.05.2025 16:45 — 👍 7 🔁 1 💬 1 📌 5
Fascinating preprint from with our "blicket detector" paradigm from Chen et al at NYU& Mila. LLM's make the same causal inference mistakes that adults make but 4 year olds don't! Of course, models are trained on adult data, kids figure it out for themselves.
im-ant.github.io/publications...
15.05.2025 18:33 — 👍 19 🔁 2 💬 2 📌 0
Professor, Department of Psychology and Center for Brain Science, Harvard University
https://gershmanlab.com/
Researching planning, reasoning, and RL in LLMs @ Reflection AI. Previously: Google DeepMind, UC Berkeley, MIT. I post about: AI 🤖, flowers 🌷, parenting 👶, public transit 🚆. She/her.
http://www.jesshamrick.com
PhD student at NYU. Interested in making machines insightful.
Cognitive scientist, philosopher, and psychologist at Berkeley, author of The Scientist in the Crib, The Philosophical Baby and The Gardener and the Carpenter and grandmother of six.
Machine Learning Professor
https://cims.nyu.edu/~andrewgw
🧙🏻♀️ scientist at Meta NYC | http://bamos.github.io
computational cog sci • problem solving and social cognition • asst prof at NYU • https://codec-lab.github.io/
Tom Griffiths' Computational Cognitive Science Lab at Princeton. Studying the computational problems human minds have to solve.
Working towards the safe development of AI for the benefit of all at Université de Montréal, LawZero and Mila.
A.M. Turing Award Recipient and most-cited AI researcher.
https://lawzero.org/en
https://yoshuabengio.org/profile/
PhD at NYU studying reasoning, decision-making, and open-endedness
alum of MIT | prev: Google, MSR, MIT CoCoSci
https://upiterbarg.github.io/
Staff ML Scientist @valenceai.bsky.social Labs/Recursion Pharma, Mila
GFlowNets, molecules & stuff
https://folinoid.com
PhD student at EPFL working on generative molecular design | Previously Microsoft AI4Science and AstraZeneca
Assistant Prof at University of Utah Fall 2025. NLP+CV+RL. RS at Google DeepMind. PhD from CMU MLD, undergrad Georgia Tech. Sometimes researcher, frequent shitposter.
Sakana AI is an AI R&D company based in Tokyo, Japan. 🗼🧠
https://sakana.ai/careers
I work at Sakana AI 🐟🐠🐡 → @sakanaai.bsky.social
https://sakana.ai/careers
|| assistant prof at University of Montreal || leading the systems neuroscience and AI lab (SNAIL: https://www.snailab.ca/) 🐌 || associate academic member of Mila (Quebec AI Institute) || #NeuroAI || vision and learning in brains and machines
Asst Professor Psychology & Data Science @ NYU | Working on brains & climate, separately | Author of Models of the Mind: How physics, engineering, and mathematics have shaped our understanding of the brain https://shorturl.at/g23c5 | Personal account (duh)
Researcher at Google and CIFAR Fellow, working on the intersection of machine learning and neuroscience in Montréal (academic affiliations: @mcgill.ca and @mila-quebec.bsky.social).
Postdoc at Meta FAIR, Comp Neuro PhD @McGill / Mila. Looking at the representation in brains and machines 🔬 https://dongyanl1n.github.io/
PhD candidate at McGill and Mila (Quebec AI Institute) w/ Blake Richards and Doina Precup.
Doing research on AI and Neuroscience 🤖🧠
Based in Montreal. 🇨🇦