@datarama joke: that's a mighty high expectation, most humans don't understand me either!
I guess you're going along the lines of the contrast with "source code can't be misunderstood by the computer, only by the humans"? I hadn't thought of that!
(In my brief interactions I don't feel particularly "misunderstood" or "understood". It just feels so random. Although the output always surprises me as more related to what I wrote than what I would expect.)
@coder That's exactly the thing, though. Humans often misunderstand me, but computers don't.
Obviously, an LLM isn't *really* misunderstanding me, because it doesn't understand anything at all. But often when I've had to interact with them I ask some question where I think it's completely obvious what I meant, and it'll generate text as though I meant something completely different.
@datarama @coder right. I'm a bit surprised because the last times I've evaluated them, I tend to resort to them when I'm stumped with something.
The generated answers are rarely correct, but frequently they are helpful (maybe just of the rubber duck effect). Which is somewhat impressive because if I was stumped, I assume I'm not asking easy questions.
(I don't think they're useful enough in the stuff I like to work on, though.)
