ICLR2023
收藏资源简介:
ICLR2023数据集由密歇根大学和北京大学共同创建,用于评估大型语言模型(LLMs)在生成高质量学术论文评审方面的能力。该数据集包含最新的开放获取研究论文及其评审,旨在解决数据污染问题,确保评估的公正性。数据集的创建过程利用了GEM(Generative Estimator for Mutual Information)方法,通过生成模型估计候选响应与参考响应之间的互信息,从而评估响应的语义信息量。ICLR2023数据集主要应用于学术评审领域,旨在提高LLMs在生成高质量评审方面的表现。
The ICLR2023 dataset, co-created by the University of Michigan and Peking University, is designed to evaluate the capacity of large language models (LLMs) in generating high-quality academic peer reviews. This dataset consists of the latest open-access research papers and their corresponding peer reviews, aiming to address data contamination issues and guarantee the fairness of evaluations. The dataset was developed using the GEM (Generative Estimator for Mutual Information) method, which estimates the mutual information between candidate responses and reference responses via generative models to assess the semantic informativeness of the responses. The ICLR2023 dataset is primarily applied in the academic peer review domain, with the goal of enhancing the performance of LLMs in generating high-quality peer reviews.




