Bergeron: Combating Adversarial Attacks through a Conscience-Based Alignment Framework

arXiv:2312.00029v2 Announce Type: replace-cross
Abstract: Research into AI alignment has grown considerably since the recent introduction of increasingly capable Large Language Models (LLMs). Unfortunately, modern methods of alignment still fail to fully prevent harmful responses when models are deliberately attacked. These attacks can trick seemingly aligned models into giving manufacturing instructions for dangerous materials, inciting violence, or recommending other immoral acts. To help mitigate this issue, we introduce Bergeron: a framework designed to improve the robustness of LLMs against attacks …

abstract adversarial adversarial attacks ai alignment alignment arxiv attacks cs.ai cs.cl cs.cr framework giving introduction language language models large language large language models llms manufacturing modern research responses through trick type