Online Rubrics Elicitation from Pairwise Comparisons

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Rezaei, MohammadHossein, Vacareanu, Robert, Wang, Zihao, Wang, Clinton, Liu, Bing, He, Yunzhong, Akyürek, Afra Feyza
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918157764001792
author Rezaei, MohammadHossein
Vacareanu, Robert
Wang, Zihao
Wang, Clinton
Liu, Bing
He, Yunzhong
Akyürek, Afra Feyza
author_facet Rezaei, MohammadHossein
Vacareanu, Robert
Wang, Zihao
Wang, Clinton
Liu, Bing
He, Yunzhong
Akyürek, Afra Feyza
contents Rubrics provide a flexible way to train LLMs on open-ended long-form answers where verifiable rewards are not applicable and human preferences provide coarse signals. Prior work shows that reinforcement learning with rubric-based rewards leads to consistent gains in LLM post-training. Most existing approaches rely on rubrics that remain static over the course of training. Such static rubrics, however, are vulnerable to reward-hacking type behaviors and fail to capture emergent desiderata that arise during training. We introduce Online Rubrics Elicitation (OnlineRubrics), a method that dynamically curates evaluation criteria in an online manner through pairwise comparisons of responses from current and reference policies. This online process enables continuous identification and mitigation of errors as training proceeds. Empirically, this approach yields consistent improvements of up to 8% over training exclusively with static rubrics across AlpacaEval, GPQA, ArenaHard as well as the validation sets of expert questions and rubrics. We qualitatively analyze the elicited criteria and identify prominent themes such as transparency, practicality, organization, and reasoning.
format Preprint
id arxiv_https___arxiv_org_abs_2510_07284
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Online Rubrics Elicitation from Pairwise Comparisons
Rezaei, MohammadHossein
Vacareanu, Robert
Wang, Zihao
Wang, Clinton
Liu, Bing
He, Yunzhong
Akyürek, Afra Feyza
Computation and Language
Artificial Intelligence
Machine Learning
Rubrics provide a flexible way to train LLMs on open-ended long-form answers where verifiable rewards are not applicable and human preferences provide coarse signals. Prior work shows that reinforcement learning with rubric-based rewards leads to consistent gains in LLM post-training. Most existing approaches rely on rubrics that remain static over the course of training. Such static rubrics, however, are vulnerable to reward-hacking type behaviors and fail to capture emergent desiderata that arise during training. We introduce Online Rubrics Elicitation (OnlineRubrics), a method that dynamically curates evaluation criteria in an online manner through pairwise comparisons of responses from current and reference policies. This online process enables continuous identification and mitigation of errors as training proceeds. Empirically, this approach yields consistent improvements of up to 8% over training exclusively with static rubrics across AlpacaEval, GPQA, ArenaHard as well as the validation sets of expert questions and rubrics. We qualitatively analyze the elicited criteria and identify prominent themes such as transparency, practicality, organization, and reasoning.
title Online Rubrics Elicitation from Pairwise Comparisons
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2510.07284