ENIGMAEVAL
收藏资源简介:
ENIGMAEVAL是一个由Scale AI发布的难题数据集,包含1184个来自拼图竞赛和事件的难题及其解决方案,旨在测试模型隐式知识综合和多步演绎推理的能力。这些难题通常需要技能娴熟的解题团队花费数小时至数天时间完成,具有明确且可验证的解决方案,便于高效评估。该数据集涵盖了文本和图像的多种非结构化格式,由全球拼图解题社区创建和分享的丰富、未充分开发的拼图库中选取。
ENIGMAEVAL is a challenging dataset released by Scale AI. It contains 1,184 puzzles and their corresponding solutions sourced from puzzle competitions and real-world incidents, and is designed to test a model's capacity for synthesizing implicit knowledge and performing multi-step deductive reasoning. These puzzles typically take skilled solving teams hours to days to complete, and feature clear and verifiable solutions that enable efficient evaluation. This dataset covers a variety of unstructured formats including text and images, and is curated from a rich, underutilized library of puzzles created and shared by the global puzzle-solving community.

- 1EnigmaEval: A Benchmark of Long Multimodal Reasoning ChallengesScale AI, Center for AI Safety, MIT · 2025年



