>composable, hierarchical generative models and probabilistic inference. Deep neural networks are basically just huge continuous circuit approximators.

These two things are not mutually exclusive.