SLMEval: Entropy-Based Calibration for Human-Aligned Evaluation of Large Language Models

Roland Daynauth, Christopher Clarke, Krisztian Flautner +2

The LLM-as-a-Judge paradigm offers a scalable, reference-free approach for evaluating language models. Although several calibration techniques have been proposed to better align th…