Rapidata/OpenAI-4o_t2i_human_preference

这是一个用于评估OpenAI 4o模型（2025年3月26日版本）在偏好、一致性和对齐性方面的数据集，包含了超过200,000个人类响应，由约45,000个注释者在不到半天的时间内使用Rapidata Python API收集。数据集包括1对1的比较，OpenAI 4o与12个其他模型进行对比，这些模型包括Ideogram V2、Recraft V2、Lumina-15-2-25等。数据集特别包含了荒谬或矛盾的提示，以测试模型在非常不可能出现在训练数据中的情境或场景下的表现。This dataset is designed to evaluate the OpenAI 4o model (version from 26.3.2025) across categories of preference, coherence, and alignment, containing over 200,000 human responses collected in less than half a day by approximately 45,000 annotators using the Rapidata Python API. The dataset includes 1v1 comparisons between OpenAI 4o and 12 other models such as Ideogram V2, Recraft V2, Lumina-15-2-25, etc. It particularly includes absurd or conflicting prompts to test the models performance in situations or scenes very unlikely to occur in the training data.