(Yes, I can see lots of ways for the idea of turning the LLM's we currently call "AI" lose to go spectacularly wrong.)