Grounding and Citations Quiz

Answers and explanations appear as you take the quiz.

1. What is the main purpose of 'grounding' an AI's answer?
   • To tie the answer to specific source material instead of memory alone
   • To make answers longer and more detailed
   • To translate the answer into multiple languages
   • To speed up how fast the model responds
2. What role do citations play in a grounded system?
   • They make the text sound more academic
   • They link claims back to the exact supporting passage
   • They hide the model's sources for privacy
   • They compress the answer to save space
3. Grounding and citations are the backbone of which common technique?
   • Gradient descent
   • Tokenization
   • Retrieval-augmented generation (RAG)
   • Batch normalization
4. If retrieved passages do not support an answer, what conservative design policy can prevent unsupported guessing?
   • The model should guess confidently anyway
   • The model should make up a citation
   • The model should return a random document
   • The model should say it could not find the answer in the sources
5. Why are ungrounded AI answers considered hard to trust for high-stakes use?
   • They are unverifiable by design, with no traceable source
   • They are always grammatically wrong
   • They take too long to generate
   • They use too much memory
6. What does an automatic 'attribution check' step do?
   • It rewrites the answer in a friendlier tone
   • It verifies each cited span actually supports the claim
   • It deletes all citations to save space
   • It translates citations into footnotes
7. What two aspects does the ALCE benchmark evaluate for long-form answers?
   • Only response length and latency
   • Audio quality and speaker identity
   • Answer quality and citation quality
   • Image resolution and color accuracy
8. Why is counting citations alone insufficient to assess grounding?
   • Every citation automatically verifies an answer
   • Grounded systems never make factual errors
   • Sources cannot contain text
   • A cited source may not actually support the attached claim