Tag
This paper argues that Generative AI evaluation should shift from static benchmarks to measuring real-world utility and human outcomes. It introduces the SCU-GenEval framework and supporting instruments to address the disconnect between benchmark performance and deployment success.