News for hackers. Stuff that matters.

A comment on Anthropic's Best-of-N Jailbreaking.

Re:Anthropic's Best-of-N Jailbreaking by infaloda on Monday December 16, 2024 @12:39AM (#42428248)
Never, have I ever read a more complicated abstract.
[ Parent | on HN ]
Re:Anthropic's Best-of-N Jailbreaking by bubblyworld on Monday December 16, 2024 @01:13AM (#42428400)
It... seems pretty ordinary to me? Like there isn't even much jargon being used. Try reading a paper in basically any field of hard science!
[ Parent | on HN ]
Re:Anthropic's Best-of-N Jailbreaking by nullandvoid on Monday December 16, 2024 @05:50AM (#42429875)
Seemed pretty simple to me and it's not my field.

My understanding was given a prompt X that is normally rejected, create Y variations with small adjustments to phrasing, grammar etc until it gives you the answer you're after.

The term "jailbreaking" used within a LLM context, is when you craft a prompt as to escape the safety sandbox, if that helps.

A sort of brute forcing the prompts if you like.

[ Parent | on HN ]