Agentic RAG Quiz

Answers and explanations appear as you take the quiz.

1. Compared with a simple single-shot retrieve-and-generate baseline, what can agentic RAG add?
   • The agent can decide whether and how to retrieve again during the task
   • Every classic RAG system makes exactly one lookup
   • Retrieval is never used
   • The model guarantees a correct answer after one search
2. How does agentic RAG handle a difficult multi-hop question?
   • It refuses to answer
   • It can break the question into sub-questions and retrieve for each
   • It always answers from memory only
   • It deletes the question
3. In agentic RAG, retrievers are best described as what to the agent?
   • Unused background processes
   • The final answer
   • Tools the agent can choose to call
   • A type of image filter
4. What does an optional 'grading' step do in agentic RAG?
   • Assigns the user a score
   • Deletes the vector store
   • Translates the answer
   • Filters out irrelevant retrieved chunks before generation
5. What is the main cost of agentic RAG compared to single-shot RAG?
   • More model calls and higher latency
   • It cannot use a vector store
   • It loses all accuracy
   • It requires no compute
6. Which scenario best fits agentic RAG?
   • Echoing back the user's exact words
   • Routing a question between internal docs, the web, and a SQL database as needed
   • Generating random text with no sources
   • Compressing an image file
7. What role do reflection tokens play in the Self-RAG framework?
   • They encrypt documents before indexing
   • They replace all retrievers with a fixed SQL query
   • They let the model signal retrieval decisions and critique retrieved evidence or generated responses
   • They guarantee that every generated citation is correct
8. How does Adaptive-RAG choose among retrieval strategies?
   • It always performs the same number of searches
   • It skips retrieval for every multi-hop question
   • It selects sources only by document length
   • It selects a retrieval approach based on estimated question complexity