PERSONA: A Reproducible Testbed for Pluralistic Alignment

TL;DR

PERSONA introduces a reproducible testbed with synthetic personas to evaluate and enhance pluralistic alignment in language models, addressing the challenge of capturing diverse user values.

Contribution

It provides a large-scale, synthetic dataset and benchmark for assessing and improving language models' ability to align with diverse user perspectives.

Findings

Generated 1,586 synthetic personas from census data.

Created a dataset with 3,868 prompts and 317,200 feedback pairs.

Established PERSONA Bench as a new benchmark for pluralistic alignment.

Abstract

The rapid advancement of language models (LMs) necessitates robust alignment with diverse user values. However, current preference optimization approaches often fail to capture the plurality of user opinions, instead reinforcing majority viewpoints and marginalizing minority perspectives. We introduce PERSONA, a reproducible test bed designed to evaluate and improve pluralistic alignment of LMs. We procedurally generate diverse user profiles from US census data, resulting in 1,586 synthetic personas with varied demographic and idiosyncratic attributes. We then generate a large-scale evaluation dataset containing 3,868 prompts and 317,200 feedback pairs obtained from our synthetic personas. Leveraging this dataset, we systematically evaluate LM capabilities in role-playing diverse users, verified through human judges, and the establishment of both a benchmark, PERSONA Bench, for pluralistic alignment approaches as well as an extensive dataset to create new and future benchmarks. The full dataset and benchmarks are available here: https://www.synthlabs.ai/research/persona.