In robotics and games, world models let agents imagine results before acting. Recent systems generate interactive video environments from prompts, and some researchers argue that learning world models — not just predicting text — is key to more general intelligence.
Whether large language models contain useful world models is an active research debate.