Re:CheatGPT
by CamperBob2 on Monday February 20, 2023 @05:54PM
(#34874006)
Welp, we're boned
[ Parent
| on HN ]
Interesting. Played around with that kind of prompt a bit and asked follow up questions on the answers as one would in a interview. The answers to those were all very vague and almost "evasive". And within 1-2 follow-ups it started to contradict itself.
I mean the first and the third aren't great. So, 33% chance of being boned.
One weird thing about ChatGPT is that it consistently feels like a B+ student, at everything you ask it. Its never top of the class, but its answers seem skilled enough to get by.
I've seen this asking about my own field (CRDTs), asking it to write comedy sketches, and with some of the academics I know asking it the sort of essay students they'd ask their students.
I wonder if its somehow doing this on purpose - like if its pitching its answers to be the most "middle of the pack" from all of its training data.
The quality of its answers improve significantly when you give it feedback and ask it to revise. It's eerie!
I feel like I could have written that. Did ChatGPT copy old blog posts of mine? ;-)
Please don't put blockquotes in code blocks.
Re:CheatGPT
by CamperBob2 on Monday February 20, 2023 @07:00PM
(#34874600)
Please ask dang (or, I suppose, pg) to implement features that were table stakes in the BBS scene in 1985.
Come to think of it, ChatGPT could probably do the job mostly by itself, given the right prompt.