Saved in:
Bibliographic Details
Main Authors: Chehbouni, Khaoula, Carr, Jonathan Colaço, More, Yash, Cheung, Jackie CK, Farnadi, Golnoosh
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2411.08243
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909635670179840
author Chehbouni, Khaoula
Carr, Jonathan Colaço
More, Yash
Cheung, Jackie CK
Farnadi, Golnoosh
author_facet Chehbouni, Khaoula
Carr, Jonathan Colaço
More, Yash
Cheung, Jackie CK
Farnadi, Golnoosh
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2411_08243
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Beyond the Safety Bundle: Auditing the Helpful and Harmless Dataset
Chehbouni, Khaoula
Carr, Jonathan Colaço
More, Yash
Cheung, Jackie CK
Farnadi, Golnoosh
Computation and Language
Computers and Society
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.
title Beyond the Safety Bundle: Auditing the Helpful and Harmless Dataset
topic Computation and Language
Computers and Society
url https://arxiv.org/abs/2411.08243