Universal and transferable adversarial attacks on aligned language models(llm-attacks.org)191 points by giuliomagnifico 3 years ago | 149 comments
