PERSONA: A Reproducible Testbed for Pluralistic Alignment
This addresses the issue of marginalizing minority perspectives in AI alignment for researchers and developers, though it is incremental as it focuses on creating a testbed rather than a new alignment method.
The authors tackled the problem of aligning language models with diverse user values by introducing PERSONA, a reproducible testbed that generates 1,586 synthetic personas and a dataset with 3,868 prompts and 317,200 feedback pairs to evaluate pluralistic alignment.
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.