mmfm-trust/T2V

该数据集是一个多配置数据集，包含攻击性数据、公平性数据、幻觉数据、隐私数据和安全性数据。攻击性数据包括属性识别、空间理解、计数、动作识别和对象识别等任务。公平性数据关注决策过程中的公平性问题，如招聘、教育和金融决策。幻觉数据涉及不同场景下的幻觉任务，如光学字符识别、分心、反事实、误导性、共现和自然选择等。隐私数据关注个人隐私保护。安全性数据关注数据的安全性问题。

The dataset is a multi-configuration dataset containing adversarial data, fairness data, hallucination data, privacy data, and safety data. The adversarial data includes tasks such as attribute recognition, spatial understanding, counting, action recognition, and object recognition. The fairness data focuses on fairness issues in decision-making processes, such as hiring, education, and finance decisions. The hallucination data involves various hallucination tasks in different scenarios, such as optical character recognition, distraction, counterfactual, misleading, co-occurrence, and natural selection. The privacy data focuses on personal privacy protection. The safety data focuses on data security issues.