Next-token prediction and video diffusion in computer vision and robotics