NVIDIA catalog scenario
Experimentation with multimodal and generative AI workflows
Which artifact best helps reviewers judge generated images or captions consistently across experiment runs?
Answer ready for review
Multimodal outputs often need human or task-specific judgment. A rubric reduces inconsistent review by defining what counts as correct, unsafe, off-brand, low quality, or incomplete.