Rich Insights from Cheap Signals: Efficient Evaluations via Tensor Factorization

TL;DR

This paper introduces a tensor factorization-based statistical model that combines inexpensive autorater data with limited human labels to enable fine-grained, efficient evaluation of generative models at the prompt level.

Contribution

A novel tensor factorization approach that effectively merges autorater scores with limited human labels for precise, scalable model evaluation and ranking.

Findings

Pretrained prompt representations improve alignment with human preferences.

The method predicts human preferences more accurately than standard baselines.

Constructs detailed prompt-based leaderboards without extensive human annotations.

Abstract

Moving beyond evaluations that collapse performance across heterogeneous prompts toward fine-grained evaluation at the prompt level, or within relatively homogeneous subsets, is necessary to diagnose generative models' strengths and weaknesses. Such fine-grained evaluations, however, suffer from a data bottleneck: human gold-standard labels are too costly at this scale, while automated ratings are often misaligned with human judgment. To resolve this challenge, we propose a novel statistical model based on tensor factorization that merges cheap autorater data with a limited set of human gold-standard labels. Specifically, our approach uses autorater scores to pretrain latent representations of prompts and generative models, and then aligns those pretrained representations to human preferences using a small calibration set. This sample-efficient methodology is robust to autorater quality, more accurately predicts human preferences on a per-prompt basis than standard baselines, and provides tight confidence intervals for key statistical parameters of interest. We also showcase the practical utility of our method by constructing granular leaderboards based on prompt qualities and by estimating model performance solely from autorater scores, eliminating the need for additional human annotations.