引導式測驗 · 中等 等級
對話測驗中的後續提示和迭代
透過有針對性的後續行動來完善回應,保留有用的工作,並在每次修訂後驗證重要的聲明和要求。
問題 1 的 8
What makes a follow-up prompt specific?
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 makes a follow-up prompt specific?
- It says only “try again”
- It restarts the task without context
- It names what should change and what should stay
- It asks for a random alternative
- When should a user restate the goal in a follow-up?
- After every single sentence
- Only when the model asks
- Never, because context cannot change
- When the task or desired outcome changes
- What should happen after a model makes a factual error?
- Assume the next answer will fix it automatically
- Provide the correction or a source and ask it to revise the affected part
- Ignore related claims
- Start over without explaining the error
- Why check important content again after a revision?
- The model may preserve an error or introduce a new one
- Follow-ups guarantee correctness
- Only first answers can contain errors
- A revision cannot change meaning
- How can a user maintain a clear task state in a long conversation?
- Repeat every prior message in full
- Assume old goals remain current
- Summarize the current goal, active constraints, decisions, and open questions
- Delete all context before each reply
- What should a user do with a calculation correction?
- Change only the final number without checking inputs
- Ask for the affected calculation to be updated and verify the result
- Assume all related sections are unchanged
- Rely on the model’s confidence
- Does a useful iterative conversation guarantee that the final answer is correct?
- Yes, if there are at least three turns
- Yes, if each follow-up is concise
- No, because follow-ups never help
- No; important claims and requirements still need review
- When can starting a fresh conversation be helpful?
- When a long thread has accumulated stale instructions or the task has changed substantially
- Whenever an answer contains a heading
- Only when the model has no context window
- Never, because summaries are impossible