NCA-GENL Trustworthy AI Practice Question
A retail company wants to let its support chatbot answer questions using internal policy documents, but executives fear the model will invent policies that do not exist. Which approach most directly reduces fabricated policy answers while keeping responses grounded in the approved documents?
⚠ Common exam trap
The trap here is equating deterministic decoding with factual accuracy, when a lower temperature only makes a fabrication repeatable rather than preventing it.
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Retrieve relevant passages from the approved policy corpus and require the model to answer only from those passages, citing them.
Grounding responses in an approved corpus through retrieval and citation is the most direct control against fabricated policies. Supplying the authoritative passages at inference time limits the model to what the documents actually say, and citations let reviewers verify each claim. Temperature, token limits, and fine-tuning change style, length, or memorized content but do not bind answers to an auditable source of truth.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Fine-tune the model on the entire policy corpus so the knowledge is baked into the weights.
Why it's wrong here
Fine-tuning can teach style and some content, but policies change frequently and weight updates are slow and costly to repeat. The model may also blend outdated and current rules, and it still lacks a mechanism to cite a specific source. Baking knowledge into weights does not prevent invention and complicates audits when a policy is revised.
- ✓
Retrieve relevant passages from the approved policy corpus and require the model to answer only from those passages, citing them.
Why this is correct
Retrieval-augmented generation restricts the context to approved passages and instructs the model to answer solely from them, which sharply reduces invention because the source of truth is supplied at inference time. Requiring citations makes each claim verifiable against the corpus, so a reviewer can immediately detect any statement not supported by a retrieved document.
- ✗
Lower the temperature to zero so the model always selects the single most probable token.
Why it's wrong here
Greedy decoding makes output more consistent but does not supply factual grounding; a model can confidently emit the same fabricated policy every time. Determinism and accuracy are different properties, and lowering temperature can even mask hallucinations by making them repeatable. Without an authoritative source in context, the model still guesses from parametric memory.
- ✗
Increase the max_tokens parameter so the model has room to explain its reasoning in full.
Why it's wrong here
Extending the output length gives the model more space to elaborate but does nothing to constrain it to source documents. A longer answer can contain just as many invented policies, simply with more words. Length limits control verbosity and cost, not factual grounding, so this does not address the executive concern about fabricated content.
About these practice questions
One of 367 original NCA-GENL practice questions on Courseiva, each with a full explanation and wrong-answer analysis — not exam dumps or protected exam content. Learn why practice questions differ from exam dumps →
JA
Written and reviewed by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
Last reviewed September 2026 · checked against the official NVIDIA exam blueprint
This NCA-GENL practice question is part of Courseiva's free NVIDIA certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the NCA-GENL exam.