GuideBench: Benchmarking Domain-Oriented Guideline Following for LLM Agents

2025-05-16

AI Summary

This paper addresses the critical challenge of large language models (LLMs) reliably adhering to dynamic, rule-dense domain-specific guidelines in domain agent scenarios. To this end, we introduce GuideBench—the first comprehensive benchmark explicitly designed to evaluate domain guideline adherence. Methodologically, we systematically construct multi-turn, updateable test cases spanning diverse domains, integrating human annotation, adversarial perturbations, and preference-based ranking evaluation. Our key contributions are threefold: (1) We formally define and quantify domain guideline adherence along three novel dimensions—rule diversity, update robustness, and human preference alignment; (2) We bridge a critical gap left by general instruction-following benchmarks in domain-specific adaptation; and (3) Through empirical evaluation, we uncover significant deficiencies in state-of-the-art LLMs—including delayed rule responsiveness and poor fine-grained compliance—providing reproducible, quantitative insights for future modeling and optimization.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Application Domains: Humanities & Computational Social Science

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Large language models (LLMs) have been widely deployed as autonomous agents capable of following user instructions and making decisions in real-world applications. Previous studies have made notable progress in benchmarking the instruction following capabilities of LLMs in general domains, with a primary focus on their inherent commonsense knowledge. Recently, LLMs have been increasingly deployed as domain-oriented agents, which rely on domain-oriented guidelines that may conflict with their commonsense knowledge. These guidelines exhibit two key characteristics: they consist of a wide range of domain-oriented rules and are subject to frequent updates. Despite these challenges, the absence of comprehensive benchmarks for evaluating the domain-oriented guideline following capabilities of LLMs presents a significant obstacle to their effective assessment and further development. In this paper, we introduce GuideBench, a comprehensive benchmark designed to evaluate guideline following performance of LLMs. GuideBench evaluates LLMs on three critical aspects: (i) adherence to diverse rules, (ii) robustness to rule updates, and (iii) alignment with human preferences. Experimental results on a range of LLMs indicate substantial opportunities for improving their ability to follow domain-oriented guidelines.

Problem

• Evaluating LLMs' adherence to domain-specific rules
• Assessing robustness to frequent guideline updates
• Measuring alignment with human preference in domains

Innovation

• Introduces GuideBench for domain-oriented LLM evaluation
• Assesses adherence, robustness, and human alignment
• Focuses on diverse, frequently updated domain rules