Tag
A research paper proposes agentic automata learning to evaluate whether LLM agents can infer hidden world models through interaction, finding that performance drops sharply as task complexity increases and that reasoning models outperform non-reasoning ones but still struggle.