Saved in:
Bibliographic Details
Main Authors: Chen, Quan Ze, Feng, K. J. Kevin, Park, Chan Young, Zhang, Amy X.
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2411.10912
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912603829174272
author Chen, Quan Ze
Feng, K. J. Kevin
Park, Chan Young
Zhang, Amy X.
author_facet Chen, Quan Ze
Feng, K. J. Kevin
Park, Chan Young
Zhang, Amy X.
contents When different groups' values differ, one approach to model alignment is to steer models at inference time towards each group's preferences. However, techniques like in-context learning only consider similarity when drawing few-shot examples and not cross-group differences in values. We propose SPICA, a framework that accounts for group-level differences during in-context example retrieval. SPICA introduces three designs: scenario banks, group-informed retrieval metrics, and in-context alignment prompts. From an evaluation of SPICA on an alignment task collecting inputs from four demographic groups ($n = 544$), our metrics retrieve in-context examples that more closely match observed preferences, with the best prompt configuration using multiple contrastive responses to demonstrate examples. In an end-to-end evaluation ($n = 120$), we observe that SPICA is higher rated than similarity-based retrieval, with groups seeing up to a +0.16 point improvement on a 5 point scale. Additionally, gains from SPICA were more uniform, with all groups benefiting from alignment rather than only some. Finally, we find that while a group-agnostic approach can align to aggregated values, it is not most suited for divergent groups.
format Preprint
id arxiv_https___arxiv_org_abs_2411_10912
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle SPICA: Retrieving Scenarios for Pluralistic In-Context Alignment
Chen, Quan Ze
Feng, K. J. Kevin
Park, Chan Young
Zhang, Amy X.
Computation and Language
When different groups' values differ, one approach to model alignment is to steer models at inference time towards each group's preferences. However, techniques like in-context learning only consider similarity when drawing few-shot examples and not cross-group differences in values. We propose SPICA, a framework that accounts for group-level differences during in-context example retrieval. SPICA introduces three designs: scenario banks, group-informed retrieval metrics, and in-context alignment prompts. From an evaluation of SPICA on an alignment task collecting inputs from four demographic groups ($n = 544$), our metrics retrieve in-context examples that more closely match observed preferences, with the best prompt configuration using multiple contrastive responses to demonstrate examples. In an end-to-end evaluation ($n = 120$), we observe that SPICA is higher rated than similarity-based retrieval, with groups seeing up to a +0.16 point improvement on a 5 point scale. Additionally, gains from SPICA were more uniform, with all groups benefiting from alignment rather than only some. Finally, we find that while a group-agnostic approach can align to aggregated values, it is not most suited for divergent groups.
title SPICA: Retrieving Scenarios for Pluralistic In-Context Alignment
topic Computation and Language
url https://arxiv.org/abs/2411.10912