I read a paper some time ago, where somebody ran a facial recognition algorithm on the output of a GAN face generator. It found lots of images in the training d

I read a paper some time ago, where somebody ran a facial recognition algorithm on the output of a GAN face generator. It found lots of images in the training data that looked strikingly similar. In other words, one reason AI images look so good, is that look a lot like actual images. Thing, is for the love of me, I can't find the paper anymore? Who knows it?