LLM-GLOBE
收藏资源简介:
LLM-GLOBE数据集由清华大学创建,旨在评估大型语言模型(LLM)输出的文化价值系统。该数据集基于GLOBE框架,包含17370条人类调查响应,涵盖62个社会和文化。数据集内容丰富,涉及九个文化价值维度,如不确定性规避、权力距离等。创建过程结合了文化心理学理论和实证验证方法,确保数据集的全球相关性和文化细微差异。该数据集主要应用于AI文化对齐研究,旨在解决AI输出与人类文化价值系统的对齐问题,促进AI技术在多元文化背景下的有效部署。
The LLM-GLOBE dataset, developed by Tsinghua University, is designed to evaluate the cultural value systems reflected in the outputs of Large Language Models (LLMs). Built on the GLOBE framework, this dataset contains 17,370 human survey responses covering 62 societies and cultures. It boasts rich content spanning nine cultural value dimensions, including uncertainty avoidance, power distance, and others. Its development integrates cultural psychology theories and empirical validation methods, ensuring the dataset's global relevance and sensitivity to subtle cultural differences. This dataset is primarily utilized in AI cultural alignment research, aiming to address the alignment problem between AI outputs and human cultural value systems, and promote the effective deployment of AI technologies in multicultural contexts.




