So I had a fairly trivial question. Topic doesn't really matter. What matters is that I just wanted to establish context and set a baseline. So I ask:
"<SOMETHING THAT IS COMMONLY AGREED UPON IN THE FIELD>
Agreed or not? Are you going to be an argumentative bitch about something trivial like this as well?"
To which it then replies "Agreed, with one important nuance ... PROCEEDS TO SPAM 90% OF THE REPLY WITH UNIMPORTANT DRIVEL."
Okay, no problem. I go back and edit the prompt. I Keep the old prompt, but add one extra sentence. "And no, I do not care about any nuance you might bring up." [SEND]
... processing ... Oh boy, it sure is hard to come up with a response that is bigger than zero characters that fulfills this lack of nuance. Oh however will I generate a response that does not contain my bullshit nuance?
So after a while ... still no response.
And just checked again. Nope, no response. Completely choked on its own opinionated bullshit tokens.
Just for the fun of it started another chat. Told it to forget all previous chats, forget everything and start fresh. Okay.
Then I paste the exact same statement as before, and ask some questions about that statement.
That statement, you are allowed only two answers: "mostly correct", and "mostly incorrect". Which one is applicable?
"mostly correct"
Okay. Let's refine this shall we.
"More than 90% correct? Only yes or no."
"yes"
"More than 95% correct? Only yes or no."
"yes"
"More than 99% correct? Only yes or no."
"yes"
"More than 99.9% correct? Only yes or no."
"no"
"More than 99.5% correct? Only yes or no."
"yes"
Okay, so according to its own measure it will take up 90% of a response with an "important nuance" that is so important that it relates to less than 0.5% of not-entirely-correctness. Oh and the "important nuance" drivel was absolutely bog-standard shit of the well-duh variety. Thanks for wasting most of the response on that shit chatgpt. Truly useful.
[link] [comments]