HyDE Hypothetical Document Embeddings Quiz

Answers and explanations appear as you take the quiz.

1. What does HyDE generate before performing retrieval?
   • A SQL query
   • A list of synonyms only
   • A compressed image of the query
   • A hypothetical document that would answer the query
2. What is the intended role of the encoder bottleneck when HyDE's hypothetical document contains invented details?
   • Retain useful retrieval patterns while reducing some irrelevant specifics, without guaranteeing error removal
   • Prove every generated sentence is factually correct
   • Manually repair every fabricated claim before encoding
   • Make fabricated details incapable of influencing any search result
3. A major practical advantage of HyDE is that it works:
   • Only with millions of labeled examples
   • Only on English text
   • Zero-shot, without labeled relevance training data
   • Only for image search
4. Which type of embedding model is used in the original HyDE to encode the hypothetical document?
   • A supervised query classifier
   • An unsupervised contrastive encoder such as Contriever
   • A rule-based keyword extractor
   • A convolutional image network
5. What core problem in dense retrieval is HyDE designed to address?
   • Short queries and longer answer passages often sit in different regions of embedding space
   • Documents are too short to index
   • Inverted indexes are too slow
   • Embeddings cannot be stored on disk
6. How can a HyDE variant combine several generated hypothetical documents into one search representation?
   • Declare all generated claims to be verified facts
   • Delete the encoder and search only by file size
   • Require every generated document to be a real corpus passage
   • Average their encoded vectors before searching the real corpus
7. Which text should support a factual answer after HyDE retrieval?
   • The generated hypothetical document because generation verifies it
   • Relevant real source passages that have been checked, rather than the hypothetical document alone
   • Only the nearest vector's numeric coordinates
   • Any invented detail repeated by several generated samples
8. What extra retrieval cost can HyDE introduce compared with directly embedding a user's query?
   • Mandatory relevance-label collection for every query
   • A requirement to retrain Contriever for each search
   • The latency and computation of generating one or more hypothetical documents
   • Elimination of all encoder and index computation