Understanding The Effect Of Temperature On Alignment With Human Opinions

TL;DR

This paper empirically compares methods for extracting human-aligned opinion distributions from large language models, highlighting the effectiveness of sampling and log-probability approaches and emphasizing the importance of understanding human subjectivity.

Contribution

It evaluates three simple methods for aligning LLM outputs with human opinions and discusses the limitations of assuming models reflect human subjectivity.

Findings

Sampling and log-probability methods outperform direct prompting in alignment.

Simple parameter adjustments improve output quality.

Assuming models mirror human opinions may be limiting.

Abstract

With the increasing capabilities of LLMs, recent studies focus on understanding whose opinions are represented by them and how to effectively extract aligned opinion distributions. We conducted an empirical analysis of three straightforward methods for obtaining distributions and evaluated the results across a variety of metrics. Our findings suggest that sampling and log-probability approaches with simple parameter adjustments can return better aligned outputs in subjective tasks compared to direct prompting. Yet, assuming models reflect human opinions may be limiting, highlighting the need for further research on how human subjectivity affects model uncertainty.