Generative AI Leader Practice Question: Techniques to Improve Generative AI Model Output
A media company uses Gemini models on Vertex AI to draft news briefs from long press releases. Editors report that drafts sometimes invent quotes and statistics that do not appear in the source. The team wants to reduce these fabrications while keeping the model's fluent writing. Which TWO techniques should they apply? (Choose two.)
⚠ Common exam trap
The trap here is treating sampling parameters such as top-k or temperature as anti-hallucination controls, when in fact raising them increases randomness and makes fabricated content more likely.
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
✓
Enable grounding with Vertex AI Search over a curated corpus of approved press releases.
Combining an explicit instruction to answer only from the supplied text with retrieval grounding over an approved corpus gives the model both a behavioral constraint and authoritative evidence. Together they reduce invented quotes and statistics while preserving fluent composition. Sampling changes like higher top-k or temperature increase randomness and work against factual fidelity, and deleting source content removes the very evidence the model needs.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Raise the top-k sampling value to let the model consider more candidate tokens.
Why it's wrong here
Top-k controls how many candidate tokens are considered at each step; increasing it widens the sampling pool and tends to increase variety and unpredictability. That can make fabrication more likely, not less, because the model is freer to choose less probable continuations. It does nothing to anchor output to the press release, so it does not address invented quotes or statistics.
- ✓
Enable grounding with Vertex AI Search over a curated corpus of approved press releases.
Why this is correct
Grounding against a curated corpus retrieves verified source passages and ties the draft to them, which directly counters invented quotes and numbers. Because the corpus is controlled, the model has authoritative evidence to draw from, and citations can be surfaced for editorial review. This complements prompt-level constraints and keeps the writing fluent while improving factual fidelity.
- ✓
Provide the press release as context and instruct the model to answer only from the provided text.
Why this is correct
Supplying the source as context and constraining the model to that text is a direct grounding instruction that reduces invented facts, because the model is told to draw only from the supplied release. It preserves fluent writing since the model still composes the brief, but it limits unsupported claims. This is a low-cost, immediate mitigation for fabricated quotes and statistics.
- ✗
Shorten the press release by removing paragraphs before sending it to the model.
Why it's wrong here
Trimming the input discards potentially relevant facts, and it does not stop the model from inventing content to fill gaps. If a removed paragraph contained the statistic being summarized, the model may hallucinate a replacement. Removing context can therefore worsen fabrication rather than reduce it, and it is not a recognized grounding technique.
- ✗
Increase the model's temperature so it paraphrases the release more creatively.
Why it's wrong here
Higher temperature increases randomness, making the model more willing to produce unlikely tokens. In a summarization task where accuracy is paramount, that raises the risk of fabricated quotes and figures rather than reducing it. Creativity is not the goal here; fidelity to the source is, so this setting works against the stated objective and should be avoided.
Go deeper
Related to this question
About these practice questions
One of 1,008 original Generative AI Leader 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 Google Cloud exam blueprint
This Generative AI Leader practice question is part of Courseiva's free Google Cloud 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 Generative AI Leader exam.