GPT-2 & human writers with a poor grasp of logic both produce content that might seem sensical on the surface, but ultimately nonsensical at a deeper level. Ho

GPT-2 & human writers with a poor grasp of logic both produce content that might seem sensical on the surface, but ultimately nonsensical at a deeper level. Hopefully we can find a way to counteract this in a systematic way. Perhaps the trick would be to punish'low order correlation' text in the first place.