Quiz#
Query Transformation#
Question 1: Why do raw user questions often yield poor search results in a RAG system?
A. The questions are too long.
B. The question’s vector does not match the vector of detailed legal documents.
C. The database is too small.
D. The system uses keyword search instead of vectors.
Answer: B
Question 2: What is the core idea of the Query Transformation technique?
A. To compress the database size.
B. To manually review user queries.
C. To use an LLM to rewrite, expand, or break down the user’s question.
D. To translate questions into multiple languages.
Answer: C
Question 3: What are the three steps of the HyDE process?
A. Encode, Decode, Generate
B. Generate, Encode, Retrieve
C. Retrieve, Read, Rank
D. Breakdown, Retrieval, Synthesis
Answer: B
Question 4: In the HyDE generation step, what does the system ask the LLM to do?
A. Write a hypothetical answer paragraph for the user’s question.
B. Search the internet for an answer.
C. Break the question into sub-queries.
D. Translate the question.
Answer: A
Question 5: What is a common characteristic of user questions compared to documents?
A. They are longer and descriptive.
B. They are identical in length.
C. They are often short and interrogative.
D. They use more complex technical vocabulary.
Answer: C
Question 6: What might happen if a user asks a very short question like ‘How to handle blue screen error’ without HyDE?
A. The system crashes.
B. The vector might mistakenly match documents describing screen colors.
C. The system automatically downloads new drivers.
D. The query is rejected.
Answer: B
Question 7: In the HyDE process, why is the ‘fake answer’ useful for retrieval?
A. Its vector is closer to the vector of the ‘real answer’ than the question’s vector.
B. It is always factually 100% accurate.
C. It bypasses the embedding model completely.
D. It relies purely on BM25 keyword matching.
Answer: A
Question 8: Which statement is true regarding the factual accuracy of the paragraph generated in the HyDE Generate step?
A. It is guaranteed to be factually perfect.
B. It must be verified by a human.
C. The information may be factually incorrect, but the style resembles the actual document.
D. It only contains information already present in the user’s prompt.
Answer: C
Question 9: When is the Query Decomposition technique particularly useful?
A. When the database is empty.
B. For simple, one-word queries.
C. For complex questions where a single text passage cannot contain enough information to answer.
D. When the user wants a hypothetical answer.
Answer: C
Question 10: What happens to a simple search vector if a question requires aggregating information from multiple sources?
A. It finds the perfect document immediately.
B. It hangs in between different topics.
C. It triggers a system timeout.
D. It switches to keyword search automatically.
Answer: B
Question 11: What is the first step in the Query Decomposition strategy?
A. Split multi-intent questions into single-intent questions.
B. Aggregate all text segments.
C. Generate a hypothetical answer.
D. Translate the query.
Answer: A
Question 12: What does the Synthesis step do in Query Decomposition?
A. Creates hypothetical answers.
B. Aggregates text segments found from all steps and gives them to the LLM.
C. Splits the question into multiple intents.
D. Generates a new vector for the question.
Answer: B
Question 13: How does Query Transformation ensure the system understands the true intent behind concise commands?
A. By skipping the embedding step.
B. By acting as an intelligent editor to edit and reorient user questions.
C. By asking the user to retype their question.
D. By only searching for the exact keywords provided.
Answer: B
Question 14: In the context of HyDE, what specifically helps the system find the exact technical instruction document for a ‘blue screen’ query?
A. The presence of technical keywords like ‘BSOD’, ‘driver’, and ‘Safe Mode’ appearing in the LLM’s draft.
B. The user explicitly typing ‘BSOD’.
C. A fallback to exact BM25 matching.
D. The decomposition of the query into three sub-queries.
Answer: A
Question 15: During Query Decomposition, why is performing a document search for each separate sub-question necessary?
A. To ensure each search has a clear goal and high accuracy.
B. To reduce the number of tokens used.
C. To bypass the vector database entirely.
D. Because the LLM cannot read long documents.
Answer: A
Question 16: If a user asks to ‘Compare the revenue of iPhone 15 and Samsung S24 in Q1 2024’, why does simple searching fail?
A. The LLM refuses to compare competitors.
B. The query length exceeds the maximum token limit.
C. There is no single document containing this comparison table, information is scattered.
D. The query lacks technical keywords.
Answer: C
Question 17: What is the ‘Encode’ step’s specific function in the HyDE process?
A. To compress the final response.
B. To pass the hypothetical paragraph through the embedding model to create a vector.
C. To encrypt user data for privacy.
D. To split the text into manageable chunks.
Answer: B
Question 18: Which technique is best to overcome semantic asymmetry by utilizing the creativity of the LLM?
A. BM25
B. Query Decomposition
C. HNSW
D. HyDE
Answer: D
Question 19: What ultimate role does Query Transformation play in the overall RAG architecture?
A. It replaces the vector database.
B. It acts as an intelligent editor, editing and reorienting questions before lookup.
C. It generates the final answer displayed to the user.
D. It handles the user interface.
Answer: B
Question 20: In a complex query scenario, what does the LLM do in the Final Generation stage of Query Decomposition?
A. It receives figures from the searches and self-aggregates them into a complete comparison answer.
B. It asks the user for clarification.
C. It generates a hypothetical document to search with.
D. It runs a Cypher query against a graph database.
Answer: A