Max Kleiman-Weiner @maxkw.bsky.social · 05/12/2024Because many of these mechanisms depend on open-ended communication, negotiation, and enforcement they have not been easy to realistically model analytically. LLM agents allow us to tackle this challenge. 100
Max Kleiman-Weiner @maxkw.bsky.social · 05/12/2024We develop GovSim, a new simulation benchmark that layers open-ended language and reasoning with a structured multi-agent MDP. 100
Max Kleiman-Weiner @maxkw.bsky.social · 05/12/2024In each GovSim challenge, five players discuss with each other, individually decide how much of each resource (e.g., fish) to extract, and then discuss the resulting outcome. After each time step the resources available double (up to a fixed capacity). 100
