ARC-AGI

该数据集是一个具有挑战性的程序合成数据集，它包含了各种独特的任务，旨在测试适应性和分布外泛化能力。此外，该基准测试还考察了开发者对先前知识的运用，以解决任务，尽管开发者无法访问私有测试集。该数据集的规模包括400个训练任务，主要针对的是程序合成任务。

This is a challenging program synthesis benchmark dataset that includes a diverse set of unique tasks, intended to evaluate adaptability and out-of-distribution (OOD) generalization capabilities. Furthermore, this benchmark assesses the utilization of prior knowledge to resolve tasks, under the condition that developers cannot access the private test set. The dataset comprises 400 training tasks, primarily targeting program synthesis tasks.