World Models' Last Exam in Physics

Video world models can generate visually convincing but physically inconsistent sequences, according to research on evaluating their reliability for prediction and planning in embodied AI systems. Existing evaluations typically depend on model-based judgments or reference videos, while direct physical tests remain largely focused elsewhere.

Video world models can produce visually convincing yet physically inconsistent sequences, raising concerns about their reliability for prediction and planning in embodied AI systems. Existing evaluations often rely on model-based judgments or reference videos, while direct physical tests largely foc