Instruction Boundary: Quantifying Biases in LLM Reasoning under Various Coverage

2025-09-24

AI Summary

This paper addresses inference biases in large language models (LLMs) arising from suboptimal prompt design—specifically, incompleteness, redundancy, or over-specification. To formalize this issue, the authors introduce the novel concept of “Instruction Boundary” and systematically define eight quantifiable dimensions characterizing prompt instruction scope. To uncover the causal relationship between prompt coverage and model reliability, they propose BiasDetector, a framework integrating controlled experiments with multi-dimensional bias detection algorithms, empirically evaluating reasoning stability and fairness across three instruction types on mainstream LLMs. Results reveal that high overall accuracy often masks substantial path-level biases, and incomplete instructions significantly increase susceptibility to adversarial manipulation in downstream tasks. This work establishes a theoretical foundation for enhancing LLM robustness and advances prompt engineering toward scientific rigor, providing a reproducible, quantitative evaluation paradigm for instruction-aware model assessment.

Technology Category

• Natural Language Processing: Prompt Engineering / Prompting
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Large-language-model (LLM) reasoning has long been regarded as a powerful tool for problem solving across domains, providing non-experts with valuable advice. However, their limitations - especially those stemming from prompt design - remain underexplored. Because users may supply biased or incomplete prompts - often unintentionally - LLMs can be misled, undermining reliability and creating risks. We refer to this vulnerability as the Instruction Boundary. To investigate the phenomenon, we distill it into eight concrete facets and introduce BiasDetector, a framework that measures biases arising from three instruction types: complete, redundant, and insufficient. We evaluate several mainstream LLMs and find that, despite high headline accuracy, substantial biases persist in many downstream tasks as a direct consequence of prompt coverage. Our empirical study confirms that LLM reasoning reliability can still be significantly improved. We analyze the practical impact of these biases and outline mitigation strategies. Our findings underscore the need for developers to tackle biases and for users to craft options carefully.

Problem

• Investigating LLM reasoning biases caused by prompt design limitations
• Measuring biases from complete, redundant, and insufficient instructions
• Analyzing how prompt coverage affects reliability in downstream tasks

Innovation

• Defining Instruction Boundary as LLM vulnerability
• Introducing BiasDetector framework for bias measurement
• Evaluating biases from three instruction coverage types