AI alignment is the project of making AI so that it won’t kill us or sell us out. A naïve conception adds a constitution or statement of principles, basically one more document to start the prompt for an LLM’s linguistic riffing. That makes the mistake of thinking that the danger is superintelligent or rogue AI, as if the AI are people who need to be raised to know the difference between right and wrong. The real danger is systems people don’t understand (used carelessly) or systems that people do understand (used maliciously). Like all problems of technology, it is a social problem that is simultaneously about people and about machines.
But really this post is just an excuse for the meme.
