問題文
Your generative AI application and your tabular model both need production quality surveillance. What distinguishes how you achieve it on each side?
選択肢
- Both use the same drift signals, since the underlying statistics are identical
- Neither needs production surveillance if the pre-deployment gate is strict enough, because a gate that covers every evaluator and every distribution check has already established the quality that production would measure
- The tabular model uses distribution-based signals and, when available, ground truth comparison; the generative application uses evaluators applied to sampled production traffic and to scheduled test datasets
- Both use evaluators, since the model monitoring service applies them to tabular models too