ChatGPT vs. the Empire State Building
I’ve been playing with ChatGPT from time to time, since I’m writing a book on Knowledge-Based AI and I need to keep track of what the Large Language Models people are up to. Here is one of the first dialogs I had with ChatGPT, from Feb 13, 2023.
(As most of you know, ChatGPT takes feedback on its answers and tries to learn from them, so this particular dialog might be much improved by now.)
So, the pumpkin and hippo answers were very good. But you’ve got pigeons, butterflies, cotton balls, cannon balls, and helium balloons, all with a terminal velocity of 150 MPH, and it missed the idea that the helium balloon would go up.
The program scolds me for cruelty to pigeons and butterflies, but not for hippos — perhaps because it “thinks” that I can’t get a hippo up there in the first place? Or maybe it just doesn’t like hippos. :-)
Admittedly, a human child might make similar errors, but would not deliver their opinions in such an authoritative and pompous manner, likely to fool people.
