WildReward: Learning Reward Models from In-the-Wild Human Interactions

TL;DR

WildReward is a novel reward model trained directly on in-the-wild human interactions, bypassing preference pairs, and demonstrating superior performance and calibration in large language model training.

Contribution

This work introduces a pipeline to extract high-quality human feedback from in-the-wild interactions for reward model training, a significant shift from preference pair reliance.

Findings

WildReward matches or outperforms traditional reward models.

It benefits from increased user diversity, enhancing reward quality.

Improves online DPO training across multiple tasks.

Abstract

Reward models (RMs) are crucial for the training of large language models (LLMs), yet they typically rely on large-scale human-annotated preference pairs. With the widespread deployment of LLMs, in-the-wild interactions have emerged as a rich source of implicit reward signals. This raises the question: Can we develop reward models directly from in-the-wild interactions? In this work, we explore this possibility by adopting WildChat as an interaction source and proposing a pipeline to extract reliable human feedback, yielding 186k high-quality instances for training WildReward via ordinal regression directly on user feedback without preference pairs. Extensive experiments demonstrate that WildReward achieves comparable or even superior performance compared to conventional reward models, with improved calibration and cross-sample consistency. We also observe that WildReward benefits directly from user diversity, where more users yield stronger reward models. Finally, we apply WildReward to online DPO training and observe significant improvements across various tasks. Code and data are released at https://github.com/THU-KEG/WildReward.