Can Language Models Serve as Text-Based World Simulators?
Dataset Analysis
LLMs are tested as text-based world simulators using a new benchmark; GPT-4 proves unreliable, highlighting limitations.
Provenance
Collected from papers.
Derived from paper: Can Language Models Serve as Text-Based World Simulators?