Guide-linked quiz · Hard Level

Keeping Chatbots On Topic Quiz

Tests scope statements, retrieval boundaries, staged controls, tool authorization, fallback behavior, and why topic checks cannot prove correctness.

Related guide pathsKeeping Chatbots On Topic
Question 1 of 8

What should a useful chatbot scope statement make clear?

Preview every question8 questions

Answers and explanations appear as you take the quiz.

  1. What should a useful chatbot scope statement make clear?
    • Only the model name
    • Topics, permitted sources and tasks, and when to refuse or hand off
    • A promise that every answer is accurate
    • The number of parameters in the model
  2. How can a team reduce unsupported answers when its approved corpus has no relevant material?
    • Hide retrieval results from evaluation
    • Require the assistant to guess
    • Provide a “not covered” response or a handoff
    • Expand tool permissions automatically
  3. Which control acts on retrieved document chunks before they enter model context?
    • Tool authorization
    • Post-release analytics
    • Retrieval check
    • Output check
  4. Why is a prompt that says “stay on topic” insufficient by itself?
    • It does not enforce data, retrieval, or tool boundaries and cannot guarantee every answer follows it
    • It prevents all user questions
    • It disables the model
    • Prompts cannot communicate any policy
  5. What should happen before a chatbot executes an external action?
    • Trust that the answer stayed on topic
    • Validate the tool and its arguments against application permissions
    • Check only the final display text
    • Retrieve more documents
  6. What does a question-to-corpus similarity threshold establish?
    • That no human handoff is needed
    • That the answer is correct
    • That a request is legally safe
    • A routing signal, not proof that the response is grounded or correct
  7. Which evaluation set best tests whether scope controls work?
    • Only spelling errors
    • Only model benchmark prompts unrelated to the service
    • Only easy in-scope requests
    • In-scope, borderline, unsupported, adversarial, and outage scenarios
  8. Why should teams track both inappropriate answers and unnecessary refusals?
    • Only refusal rate matters
    • A perfect classifier never has tradeoffs
    • Refusals are unrelated to scope
    • A control can miss out-of-scope requests or block valid edge cases

Keep testing yourself

More quizzes picked for your level.

View all quizzes