Catchy! And a really good point. Seems like there could be room for a couple of special situations with caveats though? With the GAN formulation your generator

Catchy! And a really good point. Seems like there could be room for a couple of special situations with caveats though? With the GAN formulation your generator can be practically as good as your discriminator and your discriminator can probably be better than it would have been without adversarial regularization?