Cisco Start Focus
Log In Create Account

Cisco workspace

Detailed Explanations

A detailed answer review with side-by-side rationale, distractor analysis, related questions, and weak-topic practice.

Certification
Build Quiz

Cisco catalog review

RAG

Correct: D
Your answer Not answered Answer this in practice or the daily question to sync here.
Correct answer D Retrieving relevant enterprise content and supplying it to a generative model at request time
Result Awaiting answer Synced from this browser when available.
Confidence Unset Use the practice page confidence controls to calibrate this item.

Which description best captures retrieval-augmented generation in an AI application?

A Compressing GPU tensors so a model can fit into less accelerator memory
B Replacing a language model with a rules engine for deterministic responses
C Training a model from scratch on a new corpus before each user query
D Retrieving relevant enterprise content and supplying it to a generative model at request time
1. Answer captured 2. Key checked 3. Rationale review 4. Retry weak topic

Detailed explanation

Catalog rationale

Correct answer: D

RAG is useful when the answer should be grounded in current or private content. The retrieval layer selects relevant records or chunks, and the generation layer uses that context to produce the response.

Key concept RAG

Cisco Data Center AI Infrastructure Specialist / 300-640 DCAI

Exam tip Map the requirement to the managed Cisco capability.

Eliminate services that solve infrastructure, data movement, or routing when the stem asks for AI model access or governance.

Cisco service references

01 Cisco Data Center AI Infrastructure Specialist - 300-640 DCAI 01 AI fundamentals and applications 01 Describe AI-ML workload types 01 RAG

Why the wrong answers are wrong

A

Incorrect. Tensor compression is an optimization topic, not the defining purpose of RAG.

B

Incorrect. RAG still uses a generative model; it does not replace the model with rules.

C

Incorrect. RAG does not retrain the model from scratch for each question.