AI-900 Practice Question: Describe features of generative AI workloads on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Storing model responses in a cache to retrieve them faster for repeated questions

  Why it's wrong here

  Caching model responses is purely a performance optimization that stores previously computed answers and replays them for identical or similar questions. It does not query a knowledge base at inference time, so the cached answer may become stale or factually wrong over time. RAG, in contrast, retrieves fresh documents for every request to provide up-to-date context, making caching orthogonal to the grounding purpose of RAG.

• ✓

  Retrieving relevant documents from a knowledge base to provide accurate context for LLM responses

  Why this is correct

  Retrieval-augmented generation (RAG) combines a dense retriever—typically using vector embeddings and a vector database—with a large language model. The retriever selects the most relevant document passages for the user's query, and these passages are prepended to the prompt so the model generates an answer grounded in that evidence. This dramatically reduces hallucinations, overcomes the model's training-data cutoff, and allows access to private or proprietary documents without retraining.

• ✗

  Generating random responses and selecting the most relevant using a ranker model

  Why it's wrong here

  This option describes a generate-and-rank strategy (also called best-of-n sampling), where multiple candidate responses are sampled from the LLM and then scored by a separate ranker. Retrieval-augmented generation does not randomly sample answers; it first fetches relevant documents from a knowledge base and then generates one response conditioned on that retrieved context. Because the candidates come from the model's internal knowledge alone, this approach still suffers from hallucination and knowledge cutoff, rather than grounding responses in external sources.

• ✗

  A technique for making LLM responses shorter by removing irrelevant sections

  Why it's wrong here

  This describes output truncation or summarization, which is a form of post-processing applied to an already-generated response to reduce verbosity. RAG is not about making answers shorter; it actually adds retrieved evidence into the prompt, which can make the generated response more detailed and explanatory. The goal of RAG is to improve accuracy and verifiability, not to minimize token count or remove irrelevant sections.

Go deeper

Related to this question