It would be a net win if it could normalize sending prompts instead of normal communication, which is not far in terms of useless waste of energy and space to LLM output that emulates it.