JailbreakingLanguageModels

ThoughtStorms Wiki

Using one LanguageModel to fool another one into overcoming the built in inhibitions to do evil.

Compare :

Don't Create the TormentNexus
PoisonGPT

Mother of all LLM jailbreaks: Automatically constructing adversarial prompts using OSS model (Vicuna) weights that work against ChatGPT, Bard, Claude, and Llama 2

Screenshots: Demo of response without/with jailbreak suffix

Linked thread from lead author has details/PDF https://t.co/HDzGmlNNmQ pic.twitter.com/1nBStKqXri
— Riley Goodside (@goodside) July 28, 2023

Backlinks (4 items)

AIBaiting
BadBots
NLPSquared
PoisonGPT