Yeah, I think a lot of people forget that Google (and AI research like this) pumps out a lot of work like this that shows amazing new advances. However, that doesn’t mean any of this is near ready.

Here is a 2018 paper about using world models (a concept where a model is developing an understanding of a “world”) that used it to create an interactive Doom AI model - arxiv.org/abs/1803.10122 just to show that this stuff has been in the works for a long time.