Which of the following would be a common metric for evaluating Generative AI performance?

BLEU scores for text evaluation are indeed a common metric used to assess the performance of generative AI, particularly in natural language processing tasks, such as machine translation or text generation. BLEU (Bilingual Evaluation Understudy) specifically measures the overlap between the generated text and one or more reference texts, providing a quantitative way to evaluate the quality of the generated output based on lexical similarities.

This metric is crucial in the field of generative models, as it helps researchers and developers understand how closely the AI's generated texts align with expected results. By relying on a standardized score like BLEU, it becomes easier to compare different models or iterations of a model, making it a valuable tool for continuous improvement in generative AI applications.

Other options may provide useful insights into different aspects of generative AI systems, such as user engagement or cost efficiency, but they do not serve the primary purpose of direct performance evaluation in terms of output quality.