Industrializing Prediction-Powered Inference: The GLIDE Library for Reliable GenAI and Agentic Systems Evaluation

Researchers introduced GLIDE, an open-source Python library for prediction-powered inference in GenAI and agentic system evaluation.

GLIDE unifies multiple mean-estimation estimators and sampling methods through a single API, with validation tools and guidance on method choice.

The library combines human labels with model predictions to deliver statistically valid uncertainty estimates while reducing annotation cost.

A case study showed that GLIDE can save annotations while maintaining similar precision, improving reliability for LLM and agentic evaluations.