Firm or Fickle? Evaluating Large Language Models Consistency in Sequential Interactions

2025-03-28

AI Summary

This paper addresses response inconsistency in large language models (LLMs) during multi-turn interactions in high-stakes domains. Methodologically, it (1) introduces Position-Weighted Consistency (PWC) scoring to quantify response stability across dialogue turns more precisely; (2) constructs the first cross-domain, multi-difficulty consistency benchmark dataset; and (3) proposes CARG—a confidence-aware response generation framework integrating confidence calibration with controllable decoding. Empirical evaluation demonstrates that CARG improves consistency by +28.6% in PWC score while preserving original task accuracy, thereby enhancing reliability for deployment in critical applications such as healthcare and finance.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• 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

Abstract

Large Language Models (LLMs) have shown remarkable capabilities across various tasks, but their deployment in high-stake domains requires consistent performance across multiple interaction rounds. This paper introduces a comprehensive framework for evaluating and improving LLM response consistency, making three key contributions. First, we propose a novel Position-Weighted Consistency (PWC) score that captures both the importance of early-stage stability and recovery patterns in multi-turn interactions. Second, we present a carefully curated benchmark dataset spanning diverse domains and difficulty levels, specifically designed to evaluate LLM consistency under various challenging follow-up scenarios. Third, we introduce Confidence-Aware Response Generation (CARG), a framework that significantly improves response stability by incorporating model confidence signals into the generation process. Empirical results demonstrate that CARG significantly improves response stability without sacrificing accuracy, underscoring its potential for reliable LLM deployment in critical applications.

Problem

• Evaluating LLM consistency in multi-turn interactions
• Developing metrics for stability and recovery patterns
• Improving response stability without accuracy loss

Innovation

• Position-Weighted Consistency score for multi-turn interactions
• Benchmark dataset for diverse challenging follow-up scenarios
• Confidence-Aware Response Generation framework improves stability