Language model embeddings tell a story: porn is more violent and perverse than ever. What's to blame?
Is there any reason to expect that the sharp split into two groups isn't an artifact of the method? Shouldn't all embeddings generated from a corpus of front pages over time have a Waluigi-effect-generated divide of some kind?
Is there any reason to expect that the sharp split into two groups isn't an artifact of the method? Shouldn't all embeddings generated from a corpus of front pages over time have a Waluigi-effect-generated divide of some kind?