User-Assistant Bias in LLMs
This addresses a critical issue for developers and users of LLMs in ensuring balanced conversational behaviors, though it is incremental in building on existing fine-tuning and bias detection methods.
The paper tackles the problem of user-assistant bias in LLMs, where models overly rely on their own or the user's information in multi-turn conversations, leading to stubborn or agreeable behaviors. The results show that human preference alignment increases user bias, while training on chain-of-thought reasoning decreases it, with DPO enabling bidirectional adjustment that generalizes well.
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 $\textbf{UserAssist}$, which we use to benchmark, understand and manipulate the user-assistant bias in frontier LLMs. Leveraging $\textbf{UserAssist-test}$, 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 $\textbf{UserAssist-train}$, 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.