When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Xiaomin, Yu, Zhou, Zhang, Zhiwei, Chen, Xupeng, Zhang, Ziji, Zhuang, Yingying, Sadagopan, Narayanan, Beniwal, Anurag
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912565100019712
author Li, Xiaomin
Yu, Zhou
Zhang, Zhiwei
Chen, Xupeng
Zhang, Ziji
Zhuang, Yingying
Sadagopan, Narayanan
Beniwal, Anurag
author_facet Li, Xiaomin
Yu, Zhou
Zhang, Zhiwei
Chen, Xupeng
Zhang, Ziji
Zhuang, Yingying
Sadagopan, Narayanan
Beniwal, Anurag
contents Reasoning-enhanced large language models (RLLMs), whether explicitly trained for reasoning or prompted via chain-of-thought (CoT), have achieved state-of-the-art performance on many complex reasoning tasks. However, we uncover a surprising and previously overlooked phenomenon: explicit CoT reasoning can significantly degrade instruction-following accuracy. Evaluating 15 models on two benchmarks: IFEval (with simple, rule-verifiable constraints) and ComplexBench (with complex, compositional constraints), we consistently observe performance drops when CoT prompting is applied. Through large-scale case studies and an attention-based analysis, we identify common patterns where reasoning either helps (e.g., with formatting or lexical precision) or hurts (e.g., by neglecting simple constraints or introducing unnecessary content). We propose a metric, constraint attention, to quantify model focus during generation and show that CoT reasoning often diverts attention away from instruction-relevant tokens. To mitigate these effects, we introduce and evaluate four strategies: in-context learning, self-reflection, self-selective reasoning, and classifier-selective reasoning. Our results demonstrate that selective reasoning strategies, particularly classifier-selective reasoning, can substantially recover lost performance. To our knowledge, this is the first work to systematically expose reasoning-induced failures in instruction-following and offer practical mitigation strategies.
format Preprint
id arxiv_https___arxiv_org_abs_2505_11423
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMs
Li, Xiaomin
Yu, Zhou
Zhang, Zhiwei
Chen, Xupeng
Zhang, Ziji
Zhuang, Yingying
Sadagopan, Narayanan
Beniwal, Anurag
Computation and Language
Reasoning-enhanced large language models (RLLMs), whether explicitly trained for reasoning or prompted via chain-of-thought (CoT), have achieved state-of-the-art performance on many complex reasoning tasks. However, we uncover a surprising and previously overlooked phenomenon: explicit CoT reasoning can significantly degrade instruction-following accuracy. Evaluating 15 models on two benchmarks: IFEval (with simple, rule-verifiable constraints) and ComplexBench (with complex, compositional constraints), we consistently observe performance drops when CoT prompting is applied. Through large-scale case studies and an attention-based analysis, we identify common patterns where reasoning either helps (e.g., with formatting or lexical precision) or hurts (e.g., by neglecting simple constraints or introducing unnecessary content). We propose a metric, constraint attention, to quantify model focus during generation and show that CoT reasoning often diverts attention away from instruction-relevant tokens. To mitigate these effects, we introduce and evaluate four strategies: in-context learning, self-reflection, self-selective reasoning, and classifier-selective reasoning. Our results demonstrate that selective reasoning strategies, particularly classifier-selective reasoning, can substantially recover lost performance. To our knowledge, this is the first work to systematically expose reasoning-induced failures in instruction-following and offer practical mitigation strategies.
title When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMs
topic Computation and Language
url https://arxiv.org/abs/2505.11423