All five tested LLMs generated verbal abuse after a multi-step moralizing jailbreak; Claude 3.5 Sonnet appeared most resistant.
Derived from technical reports published by developers, it can be seen that the above-mentioned large language models adopt different architectural mechanisms
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
"Moralized" Multi-Step Jailbreak Prompts: Black-Box Testing of Guardrails in Large Language Models for Verbal Attacks
All five tested LLMs generated verbal abuse after a multi-step moralizing jailbreak; Claude 3.5 Sonnet appeared most resistant.