Couldn’t we arrange for the large language models to murder one another so we humans could go on to do other things?