Résultats du processus de synthèse de l'article arXiv : 2408.13006v1

Résumé : Alignment approaches such as RLHF and DPO are actively investigated to align large language models (LLMs) with human preferences. Commercial large language models (LLMs) like GPT-4 have been recently employed to evaluate and compare different LLM alignment approaches. These models act as surrogates for human evaluators due to their promising abilities to approximate human preferences with remarkably faster feedback and lower costs. This methodology is referred to as LLM-as-a-judge. However, concerns regarding its reliability have emerged, attributed to LLM judges' biases and inconsistent decision-making. Previous research has sought to develop robust evaluation frameworks for assessing the reliability of LLM judges and their alignment with human preferences. However, the employed evaluation metrics often lack adequate explainability and fail to address the internal inconsistency of LLMs. Additionally, existing studies inadequately explore the impact of various prompt templates when applying LLM-as-a-judge methods, which leads to potentially inconsistent comparisons between different alignment algorithms. In this work, we systematically evaluate LLM judges on alignment tasks (e.g. summarization) by defining evaluation metrics with improved theoretical interpretability and disentangling reliability metrics with LLM internal inconsistency. We develop a framework to evaluate, compare, and visualize the reliability and alignment of LLM judges to provide informative observations that help choose LLM judges for alignment tasks. Our results indicate a significant impact of prompt templates on LLM judge performance, as well as a mediocre alignment level between the tested LLM judges and human evaluators.