引导式测验 · 简单 等级
人工智能系统思维测验
在整个工作流程中映射 AI 服务输入、依赖项、反馈、瓶颈、边界故障、端到端结果和版本记录。
问题 1 的 8
In an AI service, what makes a feedback loop?
预览每个问题8 问题
当您参加测验时,就会出现答案和解释。
- In an AI service, what makes a feedback loop?
- A system output changes data or behavior that later returns as input
- A model is saved to a backup drive
- Two developers review the same code
- A user closes a browser tab
- Which sequence best maps the complete path described in the guide?
- Model, then interface only
- Input collection, preprocessing, retrieval, model, tools or review, storage, and feedback
- Training data, then model weights only
- User account, billing, and logout
- Why should a team record the owner and failure behavior of each dependency?
- To increase the model’s parameter count
- To remove the need for monitoring
- To know who is responsible and what happens when a component or boundary fails
- To ensure every component uses the same programming language
- A workflow spends 40 seconds in review while retrieval and generation take 3 seconds combined. Where should a systems team investigate first?
- The model’s training date
- The logo shown in the interface
- 模型层数
- The review bottleneck and how source presentation could improve it
- Why can recommendation exposure affect later system behavior?
- User reactions to shown items can become future data
- The interface changes the model’s source code automatically
- A recommendation always reveals a user’s identity
- Exposure prevents future model updates
- Which issue can create a wrong outcome without a model crash?
- A clear ownership record
- A missing field or outdated cache at a component boundary
- A version number in a log
- A reviewed source document
- What should an end-to-end check verify?
- Only that a model endpoint returns a response
- Only that the final screen loads
- The user-visible result across the complete workflow
- Only the number of retrieval documents
- What information helps trace a regression after a system change?
- Only a screenshot of the home page
- 所有用户的列表
- A model’s marketing name
- Version information for the components involved