Video Generation Models are General-Purpose Vision Learners | Letian Wang et al. | ResearchPod