BriefGPT.xyz
Jun, 2023
酷儿人是人,首先是人:解构大型语言模型中的性取向刻板印象
Queer People are People First: Deconstructing Sexual Identity Stereotypes in Large Language Models
HTML
PDF
Harnoor Dhingra, Preetiha Jayashanker, Sayali Moghe, Emma Strubell
TL;DR
LLMs 生成的文本存在社会偏见,本文通过情感分数打分分析,证明了 LLMs 生成文本存在性少数群体偏见,并展示了一种基于 SHAP 分析的启发式方法来减轻性少数群体偏见的方法
Abstract
large language models
(
llms
) are trained primarily on minimally processed web text, which exhibits the same wide range of
social biases
he
→