A single rank-one model edit can make a safety-tuned LLM jailbreak harmful questions only when the conversation is about the poisoned concept.
Birth of a transformer: A memory viewpoint
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Concept-ROT: Poisoning Concepts in Large Language Models with Model Editing
A single rank-one model edit can make a safety-tuned LLM jailbreak harmful questions only when the conversation is about the poisoned concept.