引導式測驗 · 簡單 等級

提示測驗中的 XML 標籤和分隔符

檢查您對為什麼分隔符號將指令與內容分開、標籤如何使輸出可解析以及為什麼標籤不安全的理解程度。

問題 1 的 8

What is the main reason delimiters make prompts more reliable?

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. What is the main reason delimiters make prompts more reliable?
    • They make the prompt shorter
    • The model can tell where each part of the prompt starts and ends
    • They switch the model into a special mode
    • They reduce the price per token
  2. Do you have to use specific official tag names like <document>?
    • Yes, only names from the provider's official list work
    • Yes, and tags must be valid strict XML
    • No, there is no official list, but you should use names consistently and refer to them
    • No, and tag names should change on every request
  3. Are XML tags a reliable security boundary against prompt injection?
    • No, text inside tags can still contain injection attempts that may succeed
    • Yes, models ignore everything inside tags
    • Yes, but only if tags are nested
    • Only with the <secure> tag
  4. Why ask the model to put its final result inside <answer> tags?
    • It makes the answer more accurate by default
    • It lowers the cost of output tokens
    • It hides the answer from the user
    • It gives your code a reliable place to extract the result
  5. For prompts with long documents, where has Anthropic suggested placing the question?
    • Before the documents
    • At the end, after the documents
    • Only in the system prompt
    • Split across several messages
  6. What should you do if user-supplied text contains your own closing tag, such as </document>?
    • Leave it, since models ignore it
    • Delete the entire user input
    • Escape or remove it so the input cannot break out of its section
    • Add more opening tags
  7. Which delimiters does OpenAI's prompting guidance suggest, according to the guide?
    • Triple quotes, Markdown headings or XML
    • Only emoji markers
    • Only JSON
    • Line numbers
  8. How do tags express hierarchy, such as several documents that each have a source?
    • By using capital letters
    • By numbering each line
    • By writing everything on one line
    • By nesting, for example <document> elements inside <documents>