Research talk: NUWA: Neural visual world creation with multimodal pretraining | Microsoft Research | Podwise