This sounds incredible. Have these models effectively solved the problem of trying to use a fast-processing network to predict the world's state ahead? For example, to catch a ball?