NCA-GENL Trustworthy AI Practice Question
An enterprise is deploying an LLM-based HR assistant that answers questions about leave policies. The team wants to ensure the assistant cites the current policy document rather than relying on the model's parametric memory, which may be outdated. Which approach best supports trustworthy, verifiable answers?
⚠ Common exam trap
The trap here is assuming that fine-tuning or a strong system prompt ensures current, verifiable answers, when only retrieval can ground responses in an updatable source document.
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
✓
Use retrieval-augmented generation to fetch relevant passages from the current policy document and instruct the model to cite them in its answer.
Retrieval-augmented generation grounds answers in the current policy document and enables citations, so employees can verify the source. It also decouples knowledge updates from model retraining, letting the team refresh the index when policies change. Fine-tuning, temperature tuning, and system prompts do not provide source-backed, current answers on their own.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Lower the model's temperature to zero and rely on the model's internal knowledge of HR policies.
Why it's wrong here
Lowering temperature makes outputs more deterministic but does not update the model's knowledge or provide citations. If the parametric memory is outdated, a deterministic answer can be consistently wrong. This option addresses variability, not the core requirement for current, source-backed information.
- ✗
Fine-tune the model weekly on the latest policy PDF so the knowledge is embedded in the weights.
Why it's wrong here
Weekly fine-tuning embeds knowledge in the weights but does not provide citations, and the model may still blend outdated parametric memory with new information. It also creates operational overhead and risks catastrophic forgetting. Without retrieval, the assistant cannot point to the source document, so verifiability is not achieved.
- ✗
Add a system prompt instructing the model to always answer truthfully about leave policies.
Why it's wrong here
A system prompt can shape tone and behavior but cannot supply current policy facts or guarantee citations. The model may still hallucinate or rely on stale training data. This option relies on instruction-following rather than grounding, so it does not provide the verifiability the enterprise requires.
- ✓
Use retrieval-augmented generation to fetch relevant passages from the current policy document and instruct the model to cite them in its answer.
Why this is correct
RAG grounds the answer in the current policy document and allows the model to cite the retrieved passages, making the response verifiable. When policies change, updating the document index is faster and safer than retraining. This directly addresses the requirement that answers reflect current policy rather than outdated parametric memory.
About these practice questions
Courseiva writes every NCA-GENL question from scratch — 367 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or dumps. 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.