It's visual and obvious, dude. The 1x dog is a nightmare dog, the 4x dog is a fuzzy dog, the 16x dog is a less-fuzzy dog. But the 16x cat still has occasional spurious limbs. It's obvious that the 16x cat is a sparkly cinematic 4k-lookin' cat but there's nothing in the model to demonstrate that a 64x cat is any less likely to pop an extra leg every now and then. Photorealistic renders of things that can't exist have been a staple since Deep Dream and what's clear is that the cost-per-pixel is linear while the quality-of-massed-pixels hasn't changed appreciably. Further, that accuracy isn't even a consideration - "close-up of a short furry monster kneeling" is of a short furry monster squatting and "can it tell the difference between kneeling and squatting" is NOT a throw-away problem. More than that, it's clearly not a focus of development.