Robust LLM Performance Certification via Constrained Maximum Likelihood Estimation

TL;DR

This paper introduces a constrained maximum likelihood estimation method for accurately estimating LLM failure rates by combining human labels, judge annotations, and domain constraints, outperforming existing approaches.

Contribution

The paper presents a novel, practical constrained MLE approach that integrates multiple signals and domain knowledge for more accurate LLM failure rate estimation.

Findings

Constrained MLE outperforms state-of-the-art baselines across various settings.

The method provides more accurate and lower-variance failure rate estimates.

Empirical validation demonstrates robustness across diverse experimental regimes.

Abstract

The ability to rigorously estimate the failure rates of large language models (LLMs) is a prerequisite for their safe deployment. Currently, however, practitioners often face a tradeoff between expensive human gold standards and potentially severely-biased automatic annotation schemes such as "LLM-as-a-Judge" labeling. In this paper, we propose a new, practical, and efficient approach to LLM failure rate estimation based on constrained maximum-likelihood estimation (MLE). Our method integrates three distinct signal sources: (i) a small, high-quality human-labeled calibration set, (ii) a large corpus of LLM-judge annotations, and, most importantly, (iii) additional side information via domain-specific constraints derived from known bounds on judge performance statistics. We validate our approach through a comprehensive empirical study, benchmarking it against state-of-the-art baselines like Prediction-Powered Inference (PPI). Across diverse experimental regimes -- spanning varying judge accuracies, calibration set sizes, and LLM failure rates -- our constrained MLE consistently delivers more accurate and lower-variance estimates than existing methods. By moving beyond the "black-box" use of automated judges to a flexible framework, we provide a principled, interpretable, and scalable pathway towards LLM failure-rate certification.