MetAI 联合多所大学推出基准数据集，测试大语言模型因果推理能力

发布日期：2023-07-09    点击次数：62

品玩6月12日讯，据 huggingface 页面显示，Meta AI 近日联合香港大学以及密歇根大学发布了一款基础数据集，用来测试大语言模型的因果推理能力。

这款数据集收录了超过40万个样本，并制定了一个名为 Corr2cause 的新任务。它通过一组相关语句来确定变量之间的因果关系。研发团队目前已经通过该数据集评估了17款现有的大模型产品。

通过实验，团队评估现有的 17 个大型语言模型没有一个在这项纯粹的因果推理任务上表现良好。

","gnid":"91ee6deebaaa54079","img_data":[{"flag":2,"img":[{"desc":"","height":"337","title":"","url":"http://p1.img.360kuai.com/t018ae6a30716d7df32.jpg","width":"750"}]}],"original":0,"pat":"art_src_3,fts0,sts0","powerby":"pika","pub_time":1686558180000,"pure":"","rawurl":"http://zm.news.so.com/c491fa090f58306665df9cd98cad4030","redirect":0,"rptid":"41a9608e437a3c0d","rss_ext":[],"s":"t","src":"品玩","tag":[{"clk":"ktechnology_1:香港大学","k":"香港大学","u":""},{"clk":"ktechnology_1:密歇根大学","k":"密歇根大学","u":""}],"title":"MetAI 联合多所大学推出基准数据集，测试大语言模型因果推理能力","type":"zmt","wapurl":"http://zm.news.so.com/c491fa090f58306665df9cd98cad4030","ytag":"科技:人工智能:AI技术","zmt":{"brand":{},"cert":"优质科技领域创作者","desc":"有品好玩的科技，一切与你有关。","fans_num":9264,"id":"2991151609","is_brand":"0","name":"品玩","new_verify":"7","pic":"http://p5.img.360kuai.com/t019112a1b3e04850a2.jpg","real":1,"textimg":"http://p9.img.360kuai.com/bl/0_3/t017c4d51e87f46986f.png","verify":"0"},"zmt_status":0}","errmsg":"","errno":0}