Aligning Large Language Model Evaluators with Human Judgement: Pairwise Preference Role

Pairwise preference search improves alignment between LLM evaluators and human judgements.