视频通用模型来了！DeepMind 再证「生成即理解」，何恺明参与

TL;DR — Based on the title alone, this reports a new general-purpose video model from Google DeepMind (with Kaiming He involved) that reinforces the thesis that generative modeling itself yields understanding ("generation is understanding"). It matters as further evidence that video generation models can double as general visual understanding systems.

• The article covers a DeepMind "general/universal" video model positioned as handling broad video tasks, not a single narrow objective.
• Central claim is "生成即理解" (generation is understanding): training a generative model produces representations useful for perception/understanding tasks.
• Notable authorship signal — Kaiming He (of ResNet/MAE fame) is credited as a participant, suggesting a masked-modeling or representation-learning angle.
• Note: No article body was provided, so these points are inferred from the headline only; specific benchmarks, architecture, and results could not be verified.