引導式測驗 · 中等 等級
在 LLM 應用程式中收集使用者意見回饋測驗
負責任地收集顯式和隱式訊號,謹慎解釋它們,並在將回饋用作資料之前審查隱私。
問題 1 的 8
Which is an explicit user-feedback signal?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- Which is an explicit user-feedback signal?
- A server restart
- A model parameter count
- A rating, written comment, or direct thumbs-up/down
- A hidden network route
- Why can a “regenerate” action be ambiguous?
- It always proves the answer was false
- It is the same as written consent to share data
- It cannot be recorded
- It may indicate dissatisfaction, but the user may have another reason to request a new answer
- What does OpenAI’s API feedback-sharing guidance say about sharing feedback?
- All API content is always shared by default
- It is opt-in, and a thumbs-down submission may include the conversation and files up to that point
- Only a rating number can ever be shared
- Feedback automatically deletes the original conversation
- Why should feedback be sampled across user groups and tasks?
- People who respond may not represent all users or experiences
- Every user responds equally often
- A large volume removes all bias
- Sampling determines tokenization
- Can an unrepresentative thumbs-up sample prove overall product quality?
- Yes, if the count exceeds one hundred
- Yes, because explicit feedback is always unbiased
- No; it reflects only the sampled respondents and contexts
- No, because ratings have no value
- What should happen before edited outputs are used as evaluation labels?
- Automatic training on every edit
- Human review and checks for context, privacy, and intended meaning
- Removal of all trace context
- Publication of the full conversation
- How can an app design feedback collection with privacy in mind?
- Keep every prompt indefinitely by default
- Treat implicit actions as consent
- Send uploaded files to any reviewer
- Collect minimum context, set retention rules, and make content sharing explicit
- How should feedback fit into product evaluation?
- Complement structured tests, expert review, and safety monitoring
- Replace all offline evaluations
- Replace privacy controls
- Determine model truth from user sentiment alone