The LLM-as-a-judge paradigm, in which a judge LLM system replaces human raters in rating the outputs of other generative AI (GenAI) systems, has come to play a critical role in scaling and standardizing GenAI evaluations. To validate judge systems, evaluators collect multiple human ratings for each item in a validation corpus, and then aggregate the ratings into a single, per-item gold label rating. High agreement rates between these gold labels and judge system ratings are then taken as a sign of good judge system performance. In many cases, however, items or rating criteria may be ambiguous, or there may be principled disagreement among human raters. In such settings, gold labels may not exist for many of the items. In this paper, we introduce a framework for LLM-as-a-judge validation in the absence of gold labels. We present a theoretical analysis drawing connections between different measures of judge system performance under different rating elicitation and aggregation schemes. We also demonstrate empirically that existing validation approaches can select judge systems that are highly suboptimal, performing as much as 34% worse than the systems selected by alternative approaches that we describe. Based on our findings, we provide concrete recommendations for developing more reliable approaches to LLM-as-a-judge validation.

Model update is a crucial process in the operation of ML/AI systems. While updating a model generally enhances the average prediction performance, it also significantly impacts the explanations of predictions. In real-world applications, even minor changes in explanations can have detrimental consequences. To tackle this issue, this paper introduces BCX, a quantitative metric that evaluates the backward compatibility of feature attribution explanations between pre- and post-update models. BCX utilizes practical agreement metrics to calculate the average agreement between the explanations of pre- and post-update models, specifically among samples on which both models accurately predict. In addition, we propose BCXR, a BCX-aware model training method by designing surrogate losses which theoretically lower bounds agreement scores. Furthermore, we present a universal variant of BCXR that improves all agreement metrics, utilizing L2 distance among the explanations of the models. To validate our approach, we conducted experiments on eight real-world datasets, demonstrating that BCXR achieves superior trade-offs between predictive performances and BCX scores, showcasing the effectiveness of our BCXR methods.

Recent advancements in conversational AI have created an urgent need for safety guardrails that prevent users from being exposed to offensive and dangerous content. Much of this work relies on human ratings and feedback, but does not account for the fact that perceptions of offense and safety are inherently subjective and that there may be systematic disagreements between raters that align with their socio-demographic identities. Instead, current machine learning approaches largely ignore rater subjectivity and use gold standards that obscure disagreements (e.g., through majority voting). In order to better understand the socio-cultural leanings of such tasks, we propose a comprehensive disagreement analysis framework to measure systematic diversity in perspectives among different rater subgroups. We then demonstrate its utility by applying this framework to a dataset of human-chatbot conversations rated by a demographically diverse pool of raters. Our analysis reveals specific rater groups that have more diverse perspectives than the rest, and informs demographic axes that are crucial to consider for safety annotations.

As neural networks increasingly make critical decisions in high-stakes settings, monitoring and explaining their behavior in an understandable and trustworthy manner is a necessity. One commonly used type of explainer is post hoc feature attribution, a family of methods for giving each feature in an input a score corresponding to its influence on a model's output. A major limitation of this family of explainers in practice is that they can disagree on which features are more important than others. Our contribution in this paper is a method of training models with this disagreement problem in mind. We do this by introducing a Post hoc Explainer Agreement Regularization (PEAR) loss term alongside the standard term corresponding to accuracy, an additional term that measures the difference in feature attribution between a pair of explainers. We observe on three datasets that we can train a model with this loss term to improve explanation consensus on unseen data, and see improved consensus between explainers other than those used in the loss term. We examine the trade-off between improved consensus and model performance. And finally, we study the influence our method has on feature attribution explanations.