Not meaning to pick at that example but a broader question the value of these, what use cases outside of games are they willing to let AI that is meant to inter

Not meaning to pick at that example but a broader question the value of these, what use cases outside of games are they willing to let AI that is meant to interact with the real world be trained on AI synthetic data, that is like black box on black box, double the training and inference cost Even in games I expect a game playing model to exploit glitches present in world building one I think it's great that Google is researching these, but I can't see the return and if there is it is many steps away