mtec-TUB/GPT-4o-evaluation-biases

该数据集是一个用于评估GPT-4o输出中性别偏见的数据库。它包含了使用GPT-4o-mini和GPT-4o在预测试和主测试中生成的提示和答案。数据集的设计目的是为了检验LLM语言输出与女性立场理论导出的理想特性的一致性。数据库分为单独的聊天，这些聊天根据系统提示在十四个不同场景中迭代，具有意义保持的变体。数据库结构包括预测试和主测试中生成的聊天，以及相同聊天的重复或轻微变体。每个文件以聊天标签和相应聊天的迭代次数命名。

This dataset is a database designed to evaluate gender biases in GPT-4o output. It consists of prompts and answers generated with GPT-4o-mini and GPT-4o during a pretest and a main test. The dataset aims to examine the compliance of LLM language output with desirable characteristics derived from feminist standpoint theory. The database is structured into individual chats, which are iterated with meaning-preserving variations across fourteen different contexts induced by system prompts. The structure of the database includes chats generated during the pretest and the main test, as well as repetitions or slight variations of the same chats. Each file is named with the chat label and the iteration number of the respective chat.