Find the Intention of Instruction: Comprehensive Evaluation of Instruction Understanding for Large Language Models

2024-12-27

AI Summary

This work addresses the limited capability of large language models (LLMs) to accurately identify user intent and maintain task focus under complex, ambiguous, or adversarially perturbed instructions. To this end, we introduce IoInst—a novel benchmark that systematically defines and evaluates LLMs’ comprehension of *deep intent*, i.e., the underlying semantic goal beyond surface-level phrasing, thereby extending beyond conventional benchmarks relying on clear, unambiguous instructions. IoInst establishes an interference-resilient evaluation paradigm, incorporating multi-dimensional instruction perturbations—including nesting, logical conflict, and lexical redundancy—and combines human annotation with automated metrics for cross-model and cross-mitigation strategy robustness analysis. Experiments reveal that state-of-the-art LLMs achieve sub-60% average accuracy on IoInst, exposing critical vulnerabilities in instruction interpretation and attentional drift. Furthermore, mitigation strategies such as instruction rewriting and attention masking demonstrate measurable improvements in robustness.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Computer Vision: Large Vision Models

Application Category

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

Abstract

One of the key strengths of Large Language Models (LLMs) is their ability to interact with humans by generating appropriate responses to given instructions. This ability, known as instruction-following capability, has established a foundation for the use of LLMs across various fields and serves as a crucial metric for evaluating their performance. While numerous evaluation benchmarks have been developed, most focus solely on clear and coherent instructions. However, we have noted that LLMs can become easily distracted by instruction-formatted statements, which may lead to an oversight of their instruction comprehension skills. To address this issue, we introduce the Intention of Instruction (IoInst) benchmark. This benchmark evaluates LLMs' capacity to remain focused and understand instructions without being misled by extraneous instructions. The primary objective of this benchmark is to identify the appropriate instruction that accurately guides the generation of a given context. Our findings suggest that even recently introduced state-of-the-art models still lack instruction understanding capability. Along with the proposition of IoInst in this study, we also present broad analyses of the several strategies potentially applicable to IoInst.

Problem

• Large Language Models
• Complex Instructions
• Practical Application Efficiency

Innovation

• IoInst
• Large Language Models
• Complex Instruction Understanding