User-Assistant Bias in LLMs

2025-08-16

AI Summary

This paper identifies and systematically investigates a previously unrecognized preference bias in large language models (LLMs)—termed “user-assistant bias”—where LLMs exhibit systematic asymmetry in favoring either user or assistant utterances during multi-turn dialogues. We formally define this bias and introduce UserAssist, a benchmark dataset comprising over 8,000 dialogue turns. Through controlled ablation studies, we find that instruction tuning and human preference alignment amplify user bias, whereas reasoning-oriented training mitigates it. To address this, we propose a bidirectional bias control framework based on Direct Preference Optimization (DPO). Evaluating 52 mainstream models, we observe that commercial models exhibit strong user bias, while reasoning-focused models show significantly reduced bias. Our DPO-based method enables precise, direction-specific bias calibration on training data and demonstrates robust cross-domain generalization.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Large Multimodal Models (LMMs)
• Humans and AI: Learning Human Values and Preferences

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: Cost models of using LLMs in production systems

Abstract

Large language models (LLMs) can bias towards relying on their own or the user's information in chat history, leading to overly stubborn or agreeable behaviors in multi-turn conversations. In this paper, we formalize this model characteristic as user-assistant bias and introduce an 8k multi-turn conversation dataset $ extbf{UserAssist}$, which we use to benchmark, understand and manipulate the user-assistant bias in frontier LLMs. Leveraging $ extbf{UserAssist-test}$, we first benchmark the user-assistant bias of 26 commercial and 26 open-weight models. Commercial models show various levels of user bias. Evaluation on open-weight models reveals significant user bias in the instruction-tuned models, and weak user bias in reasoning (or reasoning-distilled) models. We then perform controlled fine-tuning experiments to pinpoint the post-training recipe contributing to these bias shifts: human preference alignment increases user bias, while training on chain-of-thought reasoning traces decreases it. Finally, we demonstrate that user-assistant bias can be bidirectionally adjusted by performing direct preference optimization (DPO) on $ extbf{UserAssist-train}$, and generalizes well to both in-domain and out-of-domain conversations. Our results provide insights into how the LLM integrates information from different sources, and also a viable way to detect and control model abnormalities.

Problem

• LLMs exhibit bias towards user or assistant information in conversations
• Benchmarking user-assistant bias across 52 commercial and open models
• Controlling bias through fine-tuning and preference optimization techniques

Innovation

• Introduced UserAssist dataset for bias benchmarking
• Used controlled fine-tuning to identify bias factors
• Applied DPO for bidirectional bias adjustment