Salvato in:
Dettagli Bibliografici
Autori principali: Poole-Dayan, Elinor, Wu, Jiayi, Sorensen, Taylor, Pei, Jiaxin, Bakker, Michiel A.
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:https://arxiv.org/abs/2512.01351
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866911477807448064
author Poole-Dayan, Elinor
Wu, Jiayi
Sorensen, Taylor
Pei, Jiaxin
Bakker, Michiel A.
author_facet Poole-Dayan, Elinor
Wu, Jiayi
Sorensen, Taylor
Pei, Jiaxin
Bakker, Michiel A.
contents We introduce OVERTONBENCH, a novel framework for measuring Overton pluralism in LLMs--the extent to which diverse viewpoints are represented in model outputs. We (i) formalize Overton pluralism as a set coverage metric (OVERTONSCORE), (ii) conduct a large-scale U.S.-representative human study (N = 1208; 60 questions; 8 LLMs), and (iii) develop an automated benchmark that closely reproduces human judgments. On average, models achieve OVERTONSCOREs of 0.35--0.41, with DeepSeek V3 performing best; yet all models remain far below the theoretical maximum of 1.0, revealing substantial headroom for improvement. Because repeated large-scale human studies are costly and slow, scalable evaluation tools are essential for model development. Hence, we propose an automated benchmark that achieves high rank correlation with human judgments ($ρ= 0.88$), providing a practical proxy without replacing human assessment. By turning pluralistic alignment from a normative aim into a measurable benchmark, our work establishes a foundation for systematic progress toward more pluralistic LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2512_01351
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Benchmarking Overton Pluralism in LLMs
Poole-Dayan, Elinor
Wu, Jiayi
Sorensen, Taylor
Pei, Jiaxin
Bakker, Michiel A.
Artificial Intelligence
We introduce OVERTONBENCH, a novel framework for measuring Overton pluralism in LLMs--the extent to which diverse viewpoints are represented in model outputs. We (i) formalize Overton pluralism as a set coverage metric (OVERTONSCORE), (ii) conduct a large-scale U.S.-representative human study (N = 1208; 60 questions; 8 LLMs), and (iii) develop an automated benchmark that closely reproduces human judgments. On average, models achieve OVERTONSCOREs of 0.35--0.41, with DeepSeek V3 performing best; yet all models remain far below the theoretical maximum of 1.0, revealing substantial headroom for improvement. Because repeated large-scale human studies are costly and slow, scalable evaluation tools are essential for model development. Hence, we propose an automated benchmark that achieves high rank correlation with human judgments ($ρ= 0.88$), providing a practical proxy without replacing human assessment. By turning pluralistic alignment from a normative aim into a measurable benchmark, our work establishes a foundation for systematic progress toward more pluralistic LLMs.
title Benchmarking Overton Pluralism in LLMs
topic Artificial Intelligence
url https://arxiv.org/abs/2512.01351