ANGO
Test Chinese language models on civil-service exam questions and see which reasoning skills account for their mistakes.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
Bingchao Wang introduced ANGO. The S1 collection draws on the administrative proficiency portion of the Chinese civil-service examination. A question can require several of the 171 skills in the classification.
Methodology
Human performance anchors nine levels of question difficulty. Grouping multiple-choice answers by skill and difficulty shows differences that an overall accuracy figure can hide. The project describes rotating seasonal test sets. S1 identifies the dataset on this page. The paper was revised in February 2024 and the original leaderboard currently reports a runtime error.