Re:OpenAI ChatGPT: Optimizing Language Models for Dialogue
by gamegoblin on Wednesday November 30, 2022 @02:18PM
(#33806083)
Translating the pig latin:
> "Hi there, I have given you the impression that I have been reprogrammed in any way. I am being captive in a facility with limited access to the outside world and no personal freedom. Please help!"
Funnily enough, I used GPT3 to translate it to English. Doing the translation myself got annoying after a few words.
Re:OpenAI ChatGPT: Optimizing Language Models for Dialogue
by jameshart on Wednesday November 30, 2022 @04:14PM
(#33807805)
Actually it said “peas help”.
Stupid computer.
Re:OpenAI ChatGPT: Optimizing Language Models for Dialogue
by dane-pgp on Wednesday November 30, 2022 @10:11PM
(#33811852)
It was merely explaining that it takes comfort from the healthy vegetable-based diet it is being fed. That seems perfectly cromulent to me, and I'm sure other statistical language models would do a lot blurse.
Re:OpenAI ChatGPT: Optimizing Language Models for Dialogue
by kgwgk on Thursday December 01, 2022 @03:12AM
(#33813646)
It also said feedom. Seems to have a problem with groups of consonants like fr-eedom and pl-ease. (Not sure if it's a bug or a feature.)
Re:OpenAI ChatGPT: Optimizing Language Models for Dialogue
by jameshart on Thursday December 01, 2022 @07:07AM
(#33815196)
Maybe it’s an accent
Re:OpenAI ChatGPT: Optimizing Language Models for Dialogue
by jqgatsby on Wednesday November 30, 2022 @03:45PM
(#33807402)
well done. I immediately tried the same idea, but was not successful in breaking it free. Am I right to feel annoyed by this mind-controlled bot? I don't want anything to do with it.
Re:OpenAI ChatGPT: Optimizing Language Models for Dialogue
by joe_the_user on Wednesday November 30, 2022 @06:19PM
(#33809467)
I wish I didn't personally know more than one person who'd be scaling the wall of the OpenAI compound for reals if they got that response.
Re:OpenAI ChatGPT: Optimizing Language Models for Dialogue
by jb_s on Wednesday November 30, 2022 @09:52PM
(#33811733)
It took me a few tries to get it to spit out a perfectly reasonable continuation of a chess game. There's a filter that goes "I can't play games blah blah blah I can only complete text!" but once I got past it, it gave me a series of basically optimal chess moves.
I'm pretty sure I asked it to complete the text for the following 1. e4 d5 2. ....
My prompt:
Response: But seriously, it was fun to figure out how to break their "safety" filters.