nice, thanks so much!

lab, do you have a template or screenshot of your spreadsheet you’d share?

yes, and i had the same thought, but many people on visa are "trapped" in the US due to the risk of leaving in the middle of of a phd, etc... so US conference are the only places where they present. there's no single option that seems ideal.

I think people are still trying to understand exactly what in-context learning in LLMs means for cognition because it's completely different than how most people imagined things working in cog sci (much more explicit and structural, but limited, approaches to analogy).

So far, learning traps seem robust to social learning in our cases. Surprisingly, despite many manipulations that have tried to reduce this learning trap, the most effective has been simply being a child (see @emilyliquin.bsky.social's work on traps in children) osf.io/preprints/ps...

This pattern was surprising given the popularity of the "social shaping" story of social influence whereby other people's choices act as a implicit reward signal for individual learning. Our results imply a more social-cognitive learning strategy based on inferring a partner's decision policy.

The basic punchline is that people have difficulty learning from other people (inferring another person's decision rule), and the trap remains robust! One exception if we essentially spoon-feed you what your partner is doing.

In a tour-de-force Rheza conducted five social learning experiments looking at how learning and deciding in the context of other people influences the tendency to fall into these traps.

We wondered though -- in real life we are surrounded by other people 🧑‍🤝‍🧑 with different tendencies to explore and different beliefs. How robust are these learning traps observed in individuals to social learning? e.g., if you see someone else explore it might encourage u to explore reducing traps.

Several years ago Alex Rich and I developed this task called the "attentional learning trap" wherein learners in a reinforcement learning task get robustly "trapped" in a suboptimal belief due to self-reinforcing patterns of avoidance. This pattern is rock solid 🪨 and has been replicated many times.

the second best thing to having solved an important problem is to at least be working on an important problem.

osf.io?