Tag
This paper investigates whether LLM agents can infer hidden world models through interaction, finding that they struggle to build stable internal models as complexity increases.