☆ Yσɠƚԋσʂ ☆ to Technology@lemmy.mlEnglish • 5 months agoNew LLM jailbreak uses models’ evaluation skills against themwww.scworld.comexternal-linkmessage-square1fedilinkarrow-up18arrow-down10
arrow-up18arrow-down1external-linkNew LLM jailbreak uses models’ evaluation skills against themwww.scworld.com☆ Yσɠƚԋσʂ ☆ to Technology@lemmy.mlEnglish • 5 months agomessage-square1fedilink
minus-square@Mixel@feddit.orglinkfedilinkEnglish1•5 months 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?