It seems like on some level GANs should already be immune to that kind of detection right? If there was that severe of a telltail sign, the adversary would pick

It seems like on some level GANs should already be immune to that kind of detection right? If there was that severe of a telltail sign, the adversary would pick up on it. It feels like these networks are optimizing for fooling us, but really they're optimizing for fooling each other, and they're good enough at detection that we get fooled too.