there are things about a chat model that you can't say about humans, like, it's not really ethical to keep a human stuffed in your pocket to be your personal assistant at your whim.

I think one of the things folks struggle with in grokking the value of these models is that we're really used to tools being like you say; they're reliable and do a thing. As though there are two states of work - perfect and useless. There are other patterns to interact with information, and this puts what we used to need humans for in a place that we can do other things with it. stuff like:

- brainstorming - rubber duck debugging - casually discussing a topic - exploring ideas / knowledge - study groups (as in, having other semi-knowledge entities around to bounce ideas off of, ask questions, etc)

when it comes to self driving cars, well, that's a bit of a different story and really is more a discussion about ethics and law and those standards. I, and others like you speak of are held of the opinion that the expectation for autonomous vehicles is a bit high given the rates of human failure, but there's plenty of arguments to be made that automating and scaling a thing means you should hold it to a higher standard anyway. I don't think there's a correct answer on this one - it's complex enough to be a battery if opinion. You mention the potential for harm, and certainly that applies here.

I'm less worried about chatgpt being wrong. Much less likely to flatten me at an intersection.