What about the self improving models? Don't those lessen the need to be imaginative? One may imagine an idea and feed it to AI, then it goes through several self-checks and improvements and one may pick something from the results, which won't close the doors of further imagining?