引導式測驗 · 中等 等級

AI撰寫的身體相機音訊警察報告測驗

測試人工智慧警察報告起草:音訊限制、轉錄錯誤、歸屬、事實驗證、審計追蹤、披露和官員

問題 1 的 8

一份音頻草稿稱,一名目擊者「指認」了一名嫌疑人,但錄音中沒有包含任何身份證明。官員該做什麼?

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。

  1. 一份音頻草稿稱,一名目擊者「指認」了一名嫌疑人,但錄音中沒有包含任何身份證明。官員該做什麼?
    • Keep it because the model produced a fluent sentence
    • Ask the model to make the claim sound more cautious
    • Remove or correct the unsupported claim and verify every statement against its source
    • Rely on the report as a second witness
  2. 為什麼隨身攝影機的音訊無法捕捉到警官的觀察結果?
    • The microphone records sound from a limited position and does not record everything the officer sees
    • Audio always includes every visual detail
    • A transcript proves the camera viewed the whole scene
    • Speech recognition reconstructs all off-camera events
  3. A transcript omits “not” from “I did not see the car.” Which risk does this illustrate?
    • Generative models cannot process negation
    • Audio is automatically inadmissible
    • The speaker identity is necessarily wrong
    • A small transcription error can reverse the meaning of a statement
  4. 哪一種審查方法最能檢查產生的警察敘述?
    • Read only the summary because it is shorter
    • Compare every factual claim with the relevant full recording and other identified sources
    • Check grammar and spelling without replaying audio
    • Ask the same model to certify the report
  5. 產品供應商報告節省了大量時間。機構還應該衡量什麼?
    • Only the number of drafts created
    • Only the model’s latency
    • Report accuracy, omissions, inventions, review time, and correction rates
    • Whether officers prefer the screen layout
  6. 審核提交後,報告處於什麼狀態?
    • An independent machine witness statement
    • A verbatim body-camera transcript by definition
    • A judicial finding
    • A report the officer adopts and remains responsible for under agency policy
  7. Why preserve draft versions and edits when policy or a legal hold requires them?
    • They can show how the draft changed and help explain or investigate a disputed statement
    • They guarantee every report is accurate
    • They replace the source video
    • They let vendors avoid audits
  8. Which safeguard reduces unsupported detail in an AI-generated narrative?
    • Prompt the model to make the story more complete
    • Link statements to audio timestamps and mark information the recording does not establish
    • Remove uncertainty words from all drafts
    • Use one report template without source references