Benchmarking Distributional Alignment of Large Language Models

TL;DR

This paper introduces a benchmark to evaluate how well large language models can align their generated opinions with specific demographic groups, addressing a complex aspect of model simulation fidelity.

Contribution

It creates a new dataset and evaluation framework that consider question domain, steering, and distribution expression, advancing understanding of distributional alignment in LMs.

Findings

LLMs can more accurately describe opinion distributions than simulate them.

Open problems remain on whether and how LMs can truly simulate human opinions.

The benchmark highlights the complexity of achieving distributional alignment.

Abstract

Language models (LMs) are increasingly used as simulacra for people, yet their ability to match the distribution of views of a specific demographic group and be \textit{distributionally aligned} remains uncertain. This notion of distributional alignment is complex, as there is significant variation in the types of attributes that are simulated. Prior works have underexplored the role of three critical variables -- the question domain, steering method, and distribution expression method -- which motivates our contribution of a benchmark explicitly addressing these dimensions. We construct a dataset expanding beyond political values, create human baselines for this task, and evaluate the extent to which an LM can align with a particular group's opinion distribution to inform design choices of such simulation systems. Our analysis reveals open problems regarding if, and how, LMs can be used to simulate humans, and that LLMs can more accurately describe the opinion distribution than simulate such distributions.