Role-Play Paradox in Large Language Models: Reasoning Performance Gains and Ethical Dilemmas

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhao, Jinman, Qian, Zifan, Cao, Linbo, Wang, Yining, Ding, Yitian, Hu, Yulan, Zhang, Zeyu, Jin, Zeyong
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913675194925056
author Zhao, Jinman
Qian, Zifan
Cao, Linbo
Wang, Yining
Ding, Yitian
Hu, Yulan
Zhang, Zeyu
Jin, Zeyong
author_facet Zhao, Jinman
Qian, Zifan
Cao, Linbo
Wang, Yining
Ding, Yitian
Hu, Yulan
Zhang, Zeyu
Jin, Zeyong
contents Role-play in large language models (LLMs) enhances their ability to generate contextually relevant and high-quality responses by simulating diverse cognitive perspectives. However, our study identifies significant risks associated with this technique. First, we demonstrate that autotuning, a method used to auto-select models' roles based on the question, can lead to the generation of harmful outputs, even when the model is tasked with adopting neutral roles. Second, we investigate how different roles affect the likelihood of generating biased or harmful content. Through testing on benchmarks containing stereotypical and harmful questions, we find that role-play consistently amplifies the risk of biased outputs. Our results underscore the need for careful consideration of both role simulation and tuning processes when deploying LLMs in sensitive or high-stakes contexts.
format Preprint
id arxiv_https___arxiv_org_abs_2409_13979
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Role-Play Paradox in Large Language Models: Reasoning Performance Gains and Ethical Dilemmas
Zhao, Jinman
Qian, Zifan
Cao, Linbo
Wang, Yining
Ding, Yitian
Hu, Yulan
Zhang, Zeyu
Jin, Zeyong
Computation and Language
Role-play in large language models (LLMs) enhances their ability to generate contextually relevant and high-quality responses by simulating diverse cognitive perspectives. However, our study identifies significant risks associated with this technique. First, we demonstrate that autotuning, a method used to auto-select models' roles based on the question, can lead to the generation of harmful outputs, even when the model is tasked with adopting neutral roles. Second, we investigate how different roles affect the likelihood of generating biased or harmful content. Through testing on benchmarks containing stereotypical and harmful questions, we find that role-play consistently amplifies the risk of biased outputs. Our results underscore the need for careful consideration of both role simulation and tuning processes when deploying LLMs in sensitive or high-stakes contexts.
title Role-Play Paradox in Large Language Models: Reasoning Performance Gains and Ethical Dilemmas
topic Computation and Language
url https://arxiv.org/abs/2409.13979