1.5 Describe Retrieval Augmented Generation (RAG) and Role of Embeddings and Vector Databases AI-TECHNICAL-PRACTITIONER Practice Questions (Page 5)

Questions 21–25

1. 21expert · hard

   A company has an LLM that answers HR policy questions. The policies change frequently, and the company also needs the model to adopt a friendly, approachable tone. The team has a limited budget and cannot retrain the model frequently. Which combination of approaches is most cost-effective?

   • AUse RAG for the tone and fine-tune for the policies.
   • BUse a very large system prompt that contains all policies and the tone instructions.
   • CUse RAG to inject the current HR policies at query time, and use a system prompt to set the friendly tone.
   • DFine-tune the model on the current policies and tone, and retrain it every time a policy changes.
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A news organization wants to build an assistant that answers questions about today's articles. The articles are published continuously throughout the day. The team has a pre-trained LLM and a vector database. Which step is essential to ensure the assistant can answer questions about an article published 10 minutes ago?

   • AEmbed the new article's text and insert the vectors into the vector database as soon as the article is published.
   • BStore the article in a relational database and rely on the LLM's built-in web browsing capability.
   • CFine-tune the LLM on the new article immediately after publication.
   • DAdd the new article's URL to the LLM's training data for the next scheduled retraining.
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A company is choosing between a vector database and a traditional relational database for its RAG system. The corpus contains 5 million short product descriptions. The team needs to find products that are semantically similar to a user's natural-language query. Which capability is the deciding factor?

   • AThe relational database can store embeddings as BLOBs and use a custom Python loop to compute similarity.
   • BThe vector database supports efficient approximate nearest-neighbor search over high-dimensional embedding vectors.
   • CThe vector database can store the original text alongside the embeddings, while a relational database cannot.
   • DThe relational database supports SQL joins, which are required for RAG retrieval.
   Reveal Solution Discussion Select an answer first
5. 24expert · hard

   A RAG system uses a vector database with an HNSW index. The team notices that query latency is high and wants to reduce it. They have 20 million vectors. Which change is most likely to reduce latency while maintaining acceptable accuracy?

   • AReduce the HNSW 'efSearch' parameter, which controls how many candidate nodes are examined during a query.
   • BIncrease the embedding dimension from 384 to 1536.
   • CSwitch from cosine similarity to Euclidean distance for the distance metric.
   • DIncrease the HNSW 'M' parameter, which controls the number of connections per node.
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   Which statement best describes how the generation stage of a RAG system uses the retrieved information?

   • AIt updates the model's weights to permanently learn the retrieved information
   • BIt incorporates the retrieved text into the prompt to produce a context-aware response
   • CIt discards the retrieved information and relies solely on the model's parametric knowledge
   • DIt performs a similarity search to find the most relevant documents
   Reveal Solution Discussion Select an answer first