Timely Events Benchmark (TiEBe)
收藏资源简介:
TiEBe数据集由坎皮纳斯州立大学和Maritaca AI联合创建,旨在评估大语言模型对全球和地区重要事件的知识掌握情况。该数据集包含11,236个问答对,数据来源于维基百科的回顾页面,涵盖了2015年至2024年间全球及多个国家(如美国、巴西、法国等)的重要事件。数据集通过自动化工具从新闻文档中生成问答对,确保数据的多样性和时效性。TiEBe主要用于评估大语言模型在持续学习和地理知识差异方面的表现,旨在解决模型在处理全球事务时的不平衡问题。
The TiEBe dataset was co-created by the University of Campinas and Maritaca AI, aiming to evaluate the knowledge of large language models (LLMs) regarding globally and regionally significant events. This dataset contains 11,236 question-answer pairs, sourced from Wikipedia's retrospective pages, covering major events worldwide and across multiple countries such as the United States, Brazil, France, etc. between 2015 and 2024. The question-answer pairs are generated from news documents via automated tools, ensuring the diversity and timeliness of the data. TiEBe is primarily used to assess the performance of LLMs in continuous learning and geographical knowledge disparities, with the goal of addressing the imbalance issue in models' handling of global affairs.

- 1TiEBe: A Benchmark for Assessing the Current Knowledge of Large Language Models坎皮纳斯州立大学 (UNICAMP), Maritaca AI · 2025年



