問題文
Which combination of practices from across the lifecycle would you present as the plan for operating a generative AI application responsibly and sustainably?
選択肢
- Infrastructure and deployments declared in templates, prompts and evaluation configuration in source control, an evaluation gate before release, staged rollout with the previous version retained, and production monitoring with continuous evaluation feeding new cases back into the gate
- A large model with a long prompt and continuous evaluation, since capability and surveillance are what matter
- Manual review of every response before it reaches the user, combined with a written record of each reviewer's decision and the reasoning behind it, since a human in the loop is the only control that reliably catches the kind of problem that no automated evaluator was ever configured to look for
- Fine-tuning for every requirement and provisioned capacity for reliability