CAVOK@lemmy.world to Technology@lemmy.ml · 1 year agoA New Attack Impacts ChatGPT—and No One Knows How to Stop Itwww.wired.comexternal-linkmessage-square3fedilinkarrow-up17arrow-down15cross-posted to: saugumas@group.ltchatgpt@lemdro.idgenerative_ai@links.hackliberty.org
arrow-up12arrow-down1external-linkA New Attack Impacts ChatGPT—and No One Knows How to Stop Itwww.wired.comCAVOK@lemmy.world to Technology@lemmy.ml · 1 year agomessage-square3fedilinkcross-posted to: saugumas@group.ltchatgpt@lemdro.idgenerative_ai@links.hackliberty.org
minus-squareKerfuffle@sh.itjust.workslinkfedilinkarrow-up9·1 year agoBy “attack” they mean “jailbreak”. It’s also nothing like a buffer overflow. The article is interesting though and the approach to generating these jailbreak prompts is creative. It looks a bit similar to the unspeakable tokens thing: https://www.vice.com/en/article/epzyva/ai-chatgpt-tokens-words-break-reddit
By “attack” they mean “jailbreak”. It’s also nothing like a buffer overflow.
The article is interesting though and the approach to generating these jailbreak prompts is creative. It looks a bit similar to the unspeakable tokens thing: https://www.vice.com/en/article/epzyva/ai-chatgpt-tokens-words-break-reddit