I predict that this will a research area that advances hugely after we solve text-to-video and text-to-3D (so maybe in 2 years?): learning these types of heuris

I predict that this will a research area that advances hugely after we solve text-to-video and text-to-3D (so maybe in 2 years?): learning these types of heuristics from billions of hours of videos. Implementing this in robots is another matter, but there has been some good progress in robotic grasping and manipulation, for example.