Beyond the Safety Bundle: Auditing the Helpful and Harmless Dataset
This work addresses dataset quality issues that can exacerbate harms in AI safety for users and developers, though it is incremental as it builds on existing auditing practices.
The study audited the Helpful and Harmless (HH) dataset used for safety mitigation in large language models, finding that conceptualization failures and quality issues lead to disparate safety behaviors across demographic groups, highlighting the need for more nuanced approaches.
In an effort to mitigate the harms of large language models (LLMs), learning from human feedback (LHF) has been used to steer LLMs towards outputs that are intended to be both less harmful and more helpful. Despite the widespread adoption of LHF in practice, the quality of this feedback and its effectiveness as a safety mitigation technique remain unclear. This study addresses these issues by auditing the widely-used Helpful and Harmless (HH) dataset by Anthropic. Our work includes: (1) a thorough investigation of the dataset's content through both manual and automated evaluation; (2) experiments demonstrating the dataset's impact on models' safety; and (3) an analysis of the 100 most influential papers citing this dataset. Through our audit, we showcase how conceptualization failures and quality issues identified in the HH dataset can create additional harms by leading to disparate safety behaviors across demographic groups. Our findings highlight the need for more nuanced, context-sensitive approaches to safety mitigation in LLMs.