I don’t think LLMs have paperclip-maximiser potential. But suppose they do. Can you unplug GPT? I can’t. To what degree do you think those who can have exploitable affections for the AI? That’s the problem.