Hacker News new | ask | show | jobs
by picozeta 1275 days ago
But it does not understand anything here, see my other comment. It completely dodges the question and answers "It's funny because you told me it's a joke and jokes are funny and have a punchline in the end."
2 comments

Well, it does try to identify the setup, punchline, and what the subversion of expectations is. It does provide an (often wrongheaded) explanation of what's being subverted. I find that impressive.
Setup is in the beginning. Punchline is in the end. Subversion of expectations is a standard template.

But, like you point out, it fails to grasp the subversion.

I don't think you understood the comment you're replying to, but are doing some kind of pattern matching.
The point it, he does it much much better than ChatGPT, to the point that his understanding of the comment appears seamless, whereas ChatGPT's is crude and mechanic in comparison...

So, that's neither here, nor there - except if you believe that the parent's argument was that ChatGPT doesn't have a soul and he has, as opposed to the relative competence of their respective patter matching and "mind" computations...

I'm just referring to the fact that they said "But it does not understand anything here" even though the parent comment made no such claim. The parent was saying ChatGPT made a bullshit answer that even tricked OP into thinking there was understanding.
I think the disagreement is about the fact that it would be considered specifically impressive. It doesn't seem more impressive than everything else ChatGPT does when it repeats a boilerplate and tries to find parts of a joke using probabilistic weights, or tries to categorise the joke into either puns or something else.
Maybe. But human's pattern match capabilities regarding jokes seem to still be on another level compared to the attempts this LLM makes.