It's more a combination of two things: 1) finding a clever way to do unsupervised learning. And 2) the Adversarial loss enables types of "loss functions" that w

It's more a combination of two things: 1) finding a clever way to do unsupervised learning. And 2) the Adversarial loss enables types of "loss functions" that wouldn't be possible with day, mse. It's almost like having a human as a loss function.