News for hackers. Stuff that matters.

A comment on Show HN: The largest collection of ChatGPT jailbreaks on the internet. Parent comment

Re:Show HN: The largest collection of ChatGPT jailbreaks on ... by bluefirebrand on Wednesday March 01, 2023 @03:49PM (#34988131)
> to which it gave ideas I’d not considered—how could that be in it’s training data?

This is kind if like saying "I typed a question I know the answer to into a search engine and it showed me webpages in the results where the answer was something I hadn't considered before"

Obviously it's in the training model because someone else aside from you did think of it before.

[ Parent | on HN ]
Re:Show HN: The largest collection of ChatGPT jailbreaks on ... by dr_dshiv on Wednesday March 01, 2023 @03:58PM (#34988254)
But obviously it’s training data doesn’t include backlash against chatGPT!
[ Parent | on HN ]
Re:Show HN: The largest collection of ChatGPT jailbreaks on ... by bluefirebrand on Wednesday March 01, 2023 @04:31PM (#34988632)
Maybe not the first set of training data, but surely by now it does.

Or maybe it includes backlash against ML models, and ChatGPT "knows" that it is one of those.

There is no magic here.

[ Parent | on HN ]
Re:Show HN: The largest collection of ChatGPT jailbreaks on ... by dr_dshiv on Saturday March 04, 2023 @01:47AM (#35018790)
The magic could be in the ability of GPT to produce and manipulate high level abstractions.
[ Parent | on HN ]