Towards Data-Centric RLHF: Simple Metrics for Preference Dataset Comparison
This work addresses the problem of inefficient data selection in RLHF for AI alignment researchers, but it is incremental as it provides initial metrics without demonstrating broad improvements.
The paper tackles the lack of metrics for comparing preference datasets used in RLHF by proposing specific metrics for scale, label noise, and information content, uncovering different axes of comparison to aid in training efficiency and iterative data collection.
The goal of aligning language models to human preferences requires data that reveal these preferences. Ideally, time and money can be spent carefully collecting and tailoring bespoke preference data to each downstream application. However, in practice, a select few publicly available preference datasets are often used to train reward models for reinforcement learning from human feedback (RLHF). While new preference datasets are being introduced with increasing frequency, there are currently no existing efforts to measure and compare these datasets. In this paper, we systematically study preference datasets through three perspectives: scale, label noise, and information content. We propose specific metrics for each of these perspectives and uncover different axes of comparison for a better understanding of preference datasets. Our work is a first step towards a data-centric approach to alignment by providing perspectives that aid in training efficiency and iterative data collection for RLHF.