XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models

arXiv:2308.01263v3 Announce Type: replace
Abstract: Without proper safeguards, large language models will readily follow malicious instructions and generate toxic content. This risk motivates safety efforts such as red-teaming and large-scale feedback learning, which aim to make models both helpful and harmless. However, there is a tension between these two objectives, since harmlessness requires models to refuse to comply with unsafe prompts, and thus not be helpful. Recent anecdotal evidence suggests that some models may have struck a poor balance, so …

abstract aim arxiv cs.ai cs.cl feedback generate however language language models large language large language models risk safeguards safety scale test type will