RigorLLM: Resilient Guardrails for Large Language Models against Undesired Content

arXiv:2403.13031v1 Announce Type: cross
Abstract: Recent advancements in Large Language Models (LLMs) have showcased remarkable capabilities across various tasks in different domains. However, the emergence of biases and the potential for generating harmful content in LLMs, particularly under malicious inputs, pose significant challenges. Current mitigation strategies, while effective, are not resilient under adversarial attacks. This paper introduces Resilient Guardrails for Large Language Models (RigorLLM), a novel framework designed to efficiently and effectively moderate harmful and unsafe inputs and outputs for …

abstract arxiv biases capabilities challenges cs.ai cs.cl cs.cr cs.lg current domains emergence guardrails however inputs language language models large language large language models llms resilient strategies tasks type