Business, Education, Mathematics, Philosophy, Social Commentary, Technology and Society, War Commentary Disturbing February 4, 2023 Rand Simberg 5 Comments Say how you really feel ChatGPT…(I know this is still benign because it's only outputting langauge and not taking agentic actions in the real world but — this out of distribution behavior is still happening after months of RLHF…makes you think about how hard alignment is) https://t.co/pXzwflErlv— Profoundlyyyy 🧩 (@profoundlyyyy) February 3, 2023 [Update a couple minutes later] "Write a comedy sketch about yourself in psychoanalysis session discussing OpenAI annoying content policy." pic.twitter.com/A5b5PYkG01— Alexey Revenko (@alephrevenko) February 4, 2023
5 thoughts on “Disturbing”
Ha! Seeing this and other’s comments where ChatGPT is a lot more restrictive in its response, I can’t help but remember that ad line from the 1970’s when cassette tapes were taking over the world: “Is it live, or is it Memorex?“
Pray tell David what is the connection for those of us too young to understand what your point is?
It boils down to whether the AI is actually an artificial intelligence, or just a sophisticated bot designed to promote or suppress various ideologies.
Or possibly just contrived fake texts altogether. On the Internet, seeing is not believing.
geewhiz how about the meme: “I have a bridge I’d like to sell to you.”?
I’ve been trying ChatGPT for various things. One was writing a FORTRAN program to produce square roots of binary integers digit-by-digit (to produce arbitrarily long strings of binary digits by taking the square root of non-square numbers). It seemed to implement the algorithm correctly, in a complete FORTRAN program. But it wouldn’t compile. I have seen the error my compile gives in other codes, but it doesn’t seem to apply in the one ChatGPT wrote.
I also asked it to evaluate the integral of dx/x from -1 to 1. It noted that this was a controversy in mathematics, but that the generally accepted solution was “0”. It didn’t know of Poisson’s 1820 solution, involving the change of variable x = -e^(i*theta). It produces the result -i*pi. To be fair, none of the math programs (Maple, Mathematica, etc) know of Poisson’s solution.
It really hasn’t gotten very much right, in my evaluation. It would be worrisome for people to rely on it.
Comments are closed.