Down the Toxicity Rabbit Hole: A Framework to Bias Audit Large Language Models with Key Emphasis on Racism, Antisemitism, and Misogyny

摘要

This paper makes three contributions. First, it presents a generalizable, novel framework dubbed toxicity rabbit hole that iteratively elicits toxic content from a wide suite of large language models. Spanning a set of 1,266 identity groups, we first conduct a bias audit of PaLM 2 guardrails presenting key insights. Next, we report generalizability across several other models. Through the elicited toxic content, we present a broad analysis with a key emphasis on racism, antisemitism, misogyny, Islamophobia, homophobia, and transphobia. We release a massive dataset of machine-generated toxic content with a view toward safety for all. Finally, driven by concrete examples, we discuss potential ramifications.