A central challenge in robotics is building policies that generalize beyond training demonstrations, which requires understanding a task's underlying physics rather than mimicking actions. The standard approach uses a vision-language-action (VLA) model, built on a pretrained vision-language model (VLM) backbone. However, a VLM learns to describe the world, not predict how it evolves, limiting physical generalization. A growing line of research replaces the language backbone with a video world model, producing a world action model (WAM). NVIDIA researcher Jim Fan explored this shift in his Robotics’ End Game talk, summarized as “VLAs are dead, long live World Action Models.” A WAM builds the policy on a video world model, so post-training specializes a model that already has a physics prior. The NVIDIA research paper World Action Models are Zero-shot Policies shows that jointly predicting video and action gives a policy properties a VLA cannot easily acquire: it learns from diverse data, as any interaction data teaches it something, and it generalizes in the open world because physics is more general than semantics. The post explores how post-training can turn WAMs into specialized robot policies, how WAMs compare to VLAs, and why the open NVIDIA Cosmos 3 world model provides a strong foundation for building WAMs.