@inthehands@hachyderm.io yeah a key thing about LLMs is the erratic behavior, so you can’t really say it always does x. It may occasionally create something new by accident, but mostly it just reconstitutes data in an arbitrary and/or verbatim way. It will sometimes regurgitate facts, but it will also sometimes describe things completely fallaciously. And it also never transforms exclusively with the specific intent to transform and create something new. It’s a bundle of contradictions.