This neural net could create a realistic 3D environment where AI agents could train and explore. The lack of embodiment is the main thing missing from reinforcement learning agents to close the gap to humans. The environment itself can be controllable by embeddings, spanning any conceivable situation. Of course, ideally, if this architecture could be made much more efficient.