Judging with Confidence: Calibrating Autoraters to Preference Distributions
This addresses the issue of subjectivity in AI alignment for researchers and practitioners, though it is incremental as it builds on existing autorater methods.
The paper tackles the problem of unreliable automated judges (autoraters) for aligning large language models with human values by proposing a framework to calibrate them to model full preference distributions, resulting in improved calibration and lower positional bias while maintaining performance on objective tasks.
The alignment of large language models (LLMs) with human values increasingly relies on using other LLMs as automated judges, or ``autoraters''. However, their reliability is limited by a foundational issue: they are trained on discrete preference labels, forcing a single ground truth onto tasks that are often subjective, ambiguous, or nuanced. We argue that a reliable autorater must learn to model the full distribution of preferences defined by a target population. In this paper, we propose a general framework for calibrating probabilistic autoraters to any given preference distribution. We formalize the problem and present two learning methods tailored to different data conditions: 1) a direct supervised fine-tuning for dense, probabilistic labels, and 2) a reinforcement learning approach for sparse, binary labels. Our empirical results show that finetuning autoraters with a distribution-matching objective leads to verbalized probability predictions that are better aligned with the target preference distribution, with improved calibration and significantly lower positional bias, all while preserving performance on objective tasks.