LLMs And The Problem Of Constructive Surprise

Large language models aren't a good method for transformative creation, but they fake it superficially well.