Simulating everything, sort of: The promise and limits of world models

World models are AI systems that build internal representations of environments and simulate future events within them, distinguishing themselves from large language models by operating in real-time with spatial understanding rather than turn-based text interactions. They aim to enable practical applications including robot training and testing, 3D asset generation, and scientific simulation, though the term remains broadly defined across the field. Some prominent AI researchers view world models as addressing limitations of large language models and representing AI's next frontier, while others see them as part of a broader diversification beyond language-focused systems. Recent commercial announcements and substantial funding rounds suggest world models are transitioning from research topics to the basis for new products and companies.