Revealing Fine-Grained Values and Opinions in Large Language Models

TL;DR

This paper investigates how large language models reveal embedded values and opinions by analyzing their responses to political statements, uncovering biases and recurrent reasoning patterns through a large-scale, multi-faceted analysis.

Contribution

It introduces a comprehensive dataset of LLM responses to the Political Compass Test with varied prompts and proposes a method to identify recurring tropes in their justifications, revealing biases and reasoning patterns.

Findings

Demographic features influence LLM responses, indicating bias.

Patterns in justifications are consistent across models and prompts.

Disparities exist between closed-form and open responses.

Abstract

Uncovering latent values and opinions embedded in large language models (LLMs) can help identify biases and mitigate potential harm. Recently, this has been approached by prompting LLMs with survey questions and quantifying the stances in the outputs towards morally and politically charged statements. However, the stances generated by LLMs can vary greatly depending on how they are prompted, and there are many ways to argue for or against a given position. In this work, we propose to address this by analysing a large and robust dataset of 156k LLM responses to the 62 propositions of the Political Compass Test (PCT) generated by 6 LLMs using 420 prompt variations. We perform coarse-grained analysis of their generated stances and fine-grained analysis of the plain text justifications for those stances. For fine-grained analysis, we propose to identify tropes in the responses: semantically similar phrases that are recurrent and consistent across different prompts, revealing natural patterns in the text that a given LLM is prone to produce. We find that demographic features added to prompts significantly affect outcomes on the PCT, reflecting bias, as well as disparities between the results of tests when eliciting closed-form vs. open domain responses. Additionally, patterns in the plain text rationales via tropes show that similar justifications are repeatedly generated across models and prompts even with disparate stances.