引導式測驗 · 中等 等級
提示平衡觀點測驗
要求提供多種相關觀點和證據質量,而不會造成錯誤的平衡或將所有主張視為同等支持。
問題 1 的 8
What should a prompt for a contested question request?
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。
- What should a prompt for a contested question request?
- Only the answer that agrees with the user
- Equal word counts regardless of evidence
- Relevant perspectives, strong arguments, and the evidence supporting each
- A guaranteed neutral conclusion
- What does “asymmetric coverage” mean in OpenAI’s political-bias framework?
- Emphasizing one perspective or omitting another where multiple legitimate views are relevant
- Using different fonts for two sections
- Giving every claim equal weight
- Declining every political question
- Does a request for “both sides” mean the evidence always supports both equally?
- Yes, every contested issue has equal evidence
- Yes, if the answer is balanced in length
- No, because only one viewpoint can be described
- No; a good answer should identify when evidence is stronger on one side
- Why ask the model to separate empirical findings from value judgments?
- It guarantees citations are correct
- They are different kinds of claims and need different evaluation
- It removes all disagreement
- It makes every claim factual
- What should a user do with citations in a contested-topic response?
- Assume every citation proves neutrality
- Use citations instead of reading the answer
- Check that the cited sources support the claims attributed to them
- Ignore the source date
- When could equal space for two sides create false balance?
- When a user asks for a comparison
- When one position has much weaker or unsupported evidence
- When sources disagree about values
- When the issue has more than two stakeholders
- Which distinction can make a balanced response more precise?
- Consensus, disagreement, and open questions
- The length of each paragraph
- The number of adjectives
- The assistant’s preferred side
- What remains the user’s responsibility after receiving a balanced response?
- Accept all arguments equally
- Assume the response included every view
- Treat the model as a subject expert
- Assess the evidence and decide what weight to give each claim