A world model is an internal representation of how the environment works, enabling a system to predict consequences of actions before taking them — a concept from robotics and RL increasingly discussed for LLMs. In model-based RL, the agent learns a world model to simulate future states and plan ahead. Recent research debates whether large language models implicitly build world models from text or merely learn surface-level correlations. The concept is central to discussions about whether AI systems can achieve genuine understanding.