☆ Yσɠƚԋσʂ ☆@lemmy.ml to Technology@lemmy.mlEnglish · 8 days agoNew LLM jailbreak uses models’ evaluation skills against themwww.scworld.comexternal-linkmessage-square1fedilinkarrow-up18arrow-down12
arrow-up16arrow-down1external-linkNew LLM jailbreak uses models’ evaluation skills against themwww.scworld.com☆ Yσɠƚԋσʂ ☆@lemmy.ml to Technology@lemmy.mlEnglish · 8 days agomessage-square1fedilink
minus-squareMixel@feddit.orglinkfedilinkEnglisharrow-up1·8 days agoI mean we are talking about a success rate of over 50% isn’t it already enough if you break it once and keep the conversion going as long as it doesn’t exceed the context limit and remembers the messages?
I mean we are talking about a success rate of over 50% isn’t it already enough if you break it once and keep the conversion going as long as it doesn’t exceed the context limit and remembers the messages?