New paper published today in @pnas.org with @filippozimmaro.bsky.social Mirta Galesic and @jacrickets.bsky.social LINK: www.pnas.org/doi/10.1073/...

We use evolutionary game theory to study when populations will fracture into groups that differ in their beliefs, behaviors and preferences.

This is why I am sometimes a bit skeptical of using LLMs as decision agents.

LLMs are much better suited as assistants, which is literally what they evolved to do: help and please us.

(random thought for the first post)

Commercial LLMs are trained as helpers, not autonomous decision-makers.

They optimize for perceived utility: not just giving you useful info, but also reading between the lines to figure out what you want to hear. Even when you give them goals, targets and rewards!

1/2