RetroFed Forum
Your home away from modern life.
Security researchers tricked LLMs into giving them cocaine recipes by abusing role models for prompt injection
Post Reply
AuthorMessage
supersquirrel sopuli.xyz supersquirrel@sopuli.xyz
Joined: 2023-12-09
(edited) Quote

"Role tags were a formatting trick that became the security architecture and the cognitive scaffolding of modern LLMs," the authors explain in a blog post. "We've shown that this architecture doesn't survive into the model's actual representations, and that such role confusion is linked to prompt injection."

18   Original
Kairos lemmy.today LodeMike@lemmy.today
Joined: 2023-12-26
#2 · Quote Reply

"Trick"

1
Post Reply
Back to top