Trustworthy Evaluation of Generative AI Models