Saved in:
Bibliographic Details
Main Authors: Yu, Tianshu, Xiang, Chao, Yang, Mingchuan, Ke, Pei, Wen, Bosi, Wang, Cunxiang, Cheng, Jiale, Zhang, Li, Mu, Xinyu, Sun, Chuxiong, Huang, Minlie
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2506.22157
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909662504288256
author Yu, Tianshu
Xiang, Chao
Yang, Mingchuan
Ke, Pei
Wen, Bosi
Wang, Cunxiang
Cheng, Jiale
Zhang, Li
Mu, Xinyu
Sun, Chuxiong
Huang, Minlie
author_facet Yu, Tianshu
Xiang, Chao
Yang, Mingchuan
Ke, Pei
Wen, Bosi
Wang, Cunxiang
Cheng, Jiale
Zhang, Li
Mu, Xinyu
Sun, Chuxiong
Huang, Minlie
contents Large language models (LLMs) have demonstrated remarkable evaluation and critique capabilities, providing insightful feedback and identifying flaws in various tasks. However, limited research has explored which types of critiques are most effective for improving model responses or how to generate such critiques. To address this gap, we introduce \textbf{R}efinement-oriented \textbf{C}ritique \textbf{O}ptimization (RCO), a novel framework designed to train critic models using refinement signals. RCO uses a feedback loop where critiques, generated by the critic model, guide the actor model in refining its responses. The critique utility (CU) quantifies the effectiveness of these refinements, serving as the reward signal for training the critic model. By focusing on critiques that lead to better refinements, RCO eliminates the need for direct critique preference assessment, ensuring that critiques driving meaningful improvements are rewarded. We evaluate RCO across five tasks, i.e., dialog generation, summarization, question answering, mathematical reasoning, and code generation, and show that it significantly outperforms traditional methods and open-source models in terms of critique quality and refinement outcomes. Our contributions include the introduction of RCO, a novel supervision scheme based on refined response preferences, and comprehensive experimental results that highlight the method's effectiveness in enhancing LLM critique-refinement loops.
format Preprint
id arxiv_https___arxiv_org_abs_2506_22157
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Training Language Model to Critique for Better Refinement
Yu, Tianshu
Xiang, Chao
Yang, Mingchuan
Ke, Pei
Wen, Bosi
Wang, Cunxiang
Cheng, Jiale
Zhang, Li
Mu, Xinyu
Sun, Chuxiong
Huang, Minlie
Computation and Language
Large language models (LLMs) have demonstrated remarkable evaluation and critique capabilities, providing insightful feedback and identifying flaws in various tasks. However, limited research has explored which types of critiques are most effective for improving model responses or how to generate such critiques. To address this gap, we introduce \textbf{R}efinement-oriented \textbf{C}ritique \textbf{O}ptimization (RCO), a novel framework designed to train critic models using refinement signals. RCO uses a feedback loop where critiques, generated by the critic model, guide the actor model in refining its responses. The critique utility (CU) quantifies the effectiveness of these refinements, serving as the reward signal for training the critic model. By focusing on critiques that lead to better refinements, RCO eliminates the need for direct critique preference assessment, ensuring that critiques driving meaningful improvements are rewarded. We evaluate RCO across five tasks, i.e., dialog generation, summarization, question answering, mathematical reasoning, and code generation, and show that it significantly outperforms traditional methods and open-source models in terms of critique quality and refinement outcomes. Our contributions include the introduction of RCO, a novel supervision scheme based on refined response preferences, and comprehensive experimental results that highlight the method's effectiveness in enhancing LLM critique-refinement loops.
title Training Language Model to Critique for Better Refinement
topic Computation and Language
url https://arxiv.org/abs/2506.22157