computer science education

Preprint: Detailed AI grading rubrics did not necessarily diagnose different learning problems

Xi Zhao, Xinyue Jiao, Zhen Xu

arXiv preprint

A preprint examining 3,041 computer-science answers found that detailed AI score dimensions overlapped strongly, while feedback identified fewer misconceptions than instructor comments. Its analysis raises a practical question: can an assessment explain a learner’s error, rather than simply produce plausible scores?