Bài kiểm tra có liên kết hướng dẫn · Trung bình Cấp độ

Gỡ lỗi một câu hỏi nhắc nhở không hoạt động

Chẩn đoán lỗi nhanh chóng bằng đường cơ sở, các thay đổi có mục tiêu, trường hợp kiểm thử đại diện và kiểm tra hồi quy trên các phiên bản.

Đường dẫn hướng dẫn liên quanGỡ lỗi lời nhắc không hoạt động
Câu hỏi 1 của 8

How should a developer begin investigating a poor prompt result?

Xem trước mọi câu hỏi8 câu hỏi

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

  1. How should a developer begin investigating a poor prompt result?
    • Describe the specific failure mode, such as missing fields or unsupported claims
    • Rewrite every instruction at once
    • Change models without testing
    • Ask for a higher confidence score
  2. Why change one prompt component at a time during debugging?
    • It guarantees perfect results
    • It avoids the need for examples
    • It helps identify which change affected the output
    • It makes model versions identical
  3. What should a prompt experiment preserve as a baseline?
    • Only the newest response
    • The original prompt and a fixed set of representative test cases
    • A different test set for every version
    • No records of prompt changes
  4. Why include edge cases in prompt tests?
    • They guarantee every prompt is safe
    • They make the test set larger without purpose
    • Only edge cases matter
    • A change may help common cases but fail on unusual inputs
  5. What can help make prompt success measurable?
    • Concrete checks such as required fields, supported citations, or task completion
    • A general request for a better answer
    • The answer’s length alone
    • A model’s self-rating
  6. If a prompt fix does not solve missing citations, what else should be inspected?
    • Only the prompt’s punctuation
    • The user’s screen brightness
    • The folder name
    • Retrieved context, tools, schema validation, and model version
  7. Why keep some evaluation examples separate from prompt tuning?
    • To increase prompt length
    • To hide prompt versions
    • To check whether improvements generalize beyond examples used to revise the prompt
    • To ensure all outputs are identical
  8. Does a prompt that passes several examples guarantee success on all inputs?
    • Yes, if it is specific
    • No; test results apply to the examples and conditions evaluated
    • Yes, if a model wrote it
    • No, because prompts cannot be tested