Automated Model Evaluation

The paper discusses the challenges in evaluating large language models, including data contamination, prompt sensitivity, and benchmark creation cost. To address these, the authors propose a lossless data compression-based evaluation…