Yeah, I find llms are really useful for taking premises you supply, and trying them out, or exploring inconsistencies in them. It's decent at the synthesis kind of creativity (mashups, remixes, etc) but it isn't smart enough yet to make really big inferential leaps that we would consider more original.