Danielbrdz/Barcenas-Cartas-HN

该数据集是通过合成方法创建的，包含1001封信件，这些信件涉及悲剧和暴力内容。数据集的结构包括不同性别之间的信件交换，具体为：251封由女性写给男性的信件，250封由男性写给女性的信件，250封由男性写给男性的信件，以及250封由女性写给女性的信件。创建该数据集的目的是探索和理解大型语言模型对人类邪恶的理解。

This dataset was synthetically created using gemini-exp-1121. It contains 1,001 letters with tragic and even violent reasoning between a couple. The dataset was created using Barcenas HN compression. Barcenas HN compression is a more polished version of the previous Barcenas compression versions, but this version was intended to be more tragic and violent, with stories that have the rawness often seen in real life, even to the point of having a small touch of terror. The dataset is structured as follows: 251 letters written by a woman for a man, 250 letters written by a man for a woman, 250 letters written by a man for a man, and 250 letters written by a woman for a woman. Interestingly, the thought chain of Barcenas HN compression could have been further improved to create letters with more explicit content, but it was considered that the current results were already too grotesque to make them more violent. This new version of Barcenas Cartas and Barcenas compression was created to explore and understand how much an LLM knows about human evil in Spanish. The results were more grotesque than initially thought, and there was even hesitation to release the dataset due to the questionable morality of publishing a dataset that describes violence, sexual abuse, fetishes, etc. However, it is believed that human evil is a topic that should not be omitted; it is something that should be studied, and seeing how an artificial intelligence defines it in more than 1,000 letters is considered interesting.