Evaluation remains one of the biggest challenges in deploying AI today, especially with the rise of generative models that produce open-ended text, images, and video.