People focusing on the use of this in video games baffles me. The point isn't that it can regenerate a videogame world, the point is that it can simulate the _

People focusing on the use of this in video games baffles me. The point isn't that it can regenerate a videogame world, the point is that it can simulate the _real world_. They're using video game footage to train it because it's cheap and easy to synthesize the data they need. This system doesn't know it's simulating a game. You can give it thousands or millions of hours of real world footage and agent input and get a simulation of the real world.