Generative AI Leader Practice Exam 64 — 20 Questions

Question 1

A media production studio wants to create short video clips from text prompts for social media content. They need a Google Cloud generative AI model that can generate videos from text. Which offering should they use?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Veo. Veo is the Google Cloud generative AI model specifically built for text-to-video generation. It enables users to create short video clips from textual descriptions, which aligns perfectly with the media studio's goal of producing social media content. Other models like Imagen and Gemini handle images or text but not video output, so Veo is the correct offering.

Answer

Chirp

Answer

Gemini

Answer

Imagen

Question 2

A university research team wants to summarize long academic papers using a Google Cloud model. They need a model that supports very large context windows so an entire paper fits in a single request. Which Gemini model characteristic should they prioritize?

Accepted Answer

A model version with a long context window, such as Gemini 1.5 Pro. Summarizing an entire academic paper in one request depends on the model's input context window. Gemini 1.5 Pro's long context capability lets the team include the full document, preserving structure and cross-references. Latency, output token limits, and language coverage affect other aspects but do not remove the need for a large input window.

Answer

A model with the largest number of supported languages

Answer

A model with the lowest latency for short prompts

Answer

A model with the highest number of output tokens per response

Question 3

A global retailer's legal team is reviewing a proposed generative AI solution that drafts personalized marketing copy for customers in the EU. The team wants to confirm whether the solution can meet the EU AI Act's transparency obligations for AI-generated content while still using Google Cloud managed services. Which approach best satisfies the transparency requirement?

Accepted Answer

Use SynthID watermarking on generated images and add clear AI-generated labels to text and media outputs.. The EU AI Act requires that AI-generated content be clearly identifiable as such. SynthID watermarking provides a durable, machine-detectable signal embedded in generated media, while explicit labels inform human readers. Using these together on Vertex AI and Gemini outputs delivers the required transparency without blocking the personalization workflow or forcing impractical manual review of every message.

Answer

Log all prompts and model responses in Cloud Logging and retain the logs for twelve months.

Answer

Require human review of every generated marketing message before it is sent to customers.

Answer

Restrict the solution to internal employees so that no customer-facing AI-generated content is produced.

Question 4

A data science team is building a document question-answering assistant on Vertex AI. Users report that answers are sometimes fabricated when the retrieved passages do not contain the answer. Which TWO techniques should the team apply to reduce hallucinations? (Choose two.)

Accepted Answer

Instruct the model to answer only from the provided context and to respond with a fixed phrase such as 'not found in the provided documents' when the context is insufficient.. Grounding instructions with an abstention path and mandatory citations both push the model to rely on retrieved passages and to expose when evidence is missing. Higher temperature and larger output limits increase unsupported text, while fine-tuning stores a stale snapshot without providing query-time provenance. Together, the two prompt and output controls reduce fabrication.

Answer

Raise the maximum output token limit so the model has more space to explain its reasoning and avoid mistakes.

Answer

Increase the temperature so the model explores alternative answers when the retrieved context seems incomplete.

Answer

Fine-tune the model on the full document corpus so all answers are stored in the model weights.

Question 5

A financial services firm uses a generative AI model on Vertex AI to answer employees' HR policy questions. The model sometimes invents policy details. The firm wants answers grounded in the official HR handbook and needs to cite the source section. Which solution should they implement?

Accepted Answer

Use retrieval-augmented generation with Vertex AI Search over the HR handbook and instruct the model to cite retrieved sections.. Retrieval-augmented generation over the HR handbook supplies authoritative passages and enables citations, keeping answers grounded and current. Fine-tuning embeds knowledge but does not guarantee citation or easy updates. Low temperature does not add missing knowledge, and stuffing the full handbook into the prompt is inefficient and less precise than retrieval.

Answer

Fine-tune the model on the HR handbook text so the knowledge is embedded in the weights.

Answer

Increase the context window by using a model with a larger token limit and paste the entire handbook into every prompt.

Answer

Lower the temperature to zero and rely on the model's pretrained knowledge of HR policies.

Question 6

A media company wants its internal knowledge assistant to answer employee questions using the company's own policy documents and past project reports, while keeping the Gemini model's general reasoning ability intact. The team has a large corpus stored in Cloud Storage and does not want to retrain or fine-tune the model. Which Google Cloud approach should they use?

Accepted Answer

Use Vertex AI Search with a data store pointing to the Cloud Storage corpus, and ground Gemini responses on the retrieved documents.. Grounding Gemini with Vertex AI Search connects the model to the company's own documents at query time, so responses reflect current policy and project material while the base model's general capabilities remain untouched. Because retrieval is separate from model weights, updating the corpus requires no retraining, which matches the team's stated constraint.

Answer

Deploy Gemini on a larger machine type with more GPU memory so the model can ingest the full document corpus at runtime.

Answer

Increase the Gemini model's temperature setting so it can explore the policy documents more broadly at inference time.

Answer

Fine-tune Gemini on the policy documents and past project reports using Vertex AI supervised tuning.

Question 7

A marketing team is using a text generation model to create ad copy. They notice that the model's output is often bland and lacks creativity. They want to increase the diversity of the generated text while keeping it relevant. Which parameter should they adjust?

Accepted Answer

Temperature. Temperature is the primary parameter that controls randomness in generation. Increasing it allows the model to explore less likely token sequences, resulting in more varied and creative outputs. Other parameters like max output tokens, top-k, and frequency penalty address different aspects such as length, token pool restriction, and repetition, respectively.

Answer

Top-k

Answer

Frequency penalty

Answer

Max output tokens

Question 8

A national retail chain wants to add a generative AI shopping assistant to its existing mobile app. The CIO insists the project show measurable business value within one quarter and that spending stay predictable, with no long-term infrastructure commitments. Which Google Cloud approach best fits these constraints?

Accepted Answer

Consume Gemini models through Vertex AI with pay-as-you-go pricing and ground responses in the retailer's product catalog.. Managed access to Gemini models on Vertex AI removes infrastructure ownership and converts spending into usage-based costs, matching the demand for predictable budgets. Grounding in the retailer's product catalog raises answer quality so the assistant delivers visible shopper value quickly. Together they let the retailer launch within the quarter instead of funding a lengthy custom modeling effort.

Answer

Deploy an open-weight model on Compute Engine VMs sized for peak holiday traffic and manage scaling manually.

Answer

Purchase dedicated TPU capacity in a specific region and build a custom model training pipeline before any customer-facing launch.

Answer

Build a bespoke large language model from scratch using the retailer's historical chat transcripts as the only training data.

Question 9

A media company is using a generative AI model to create summaries of news articles. They notice that the summaries sometimes include information not present in the original articles, leading to factual errors. They want to reduce the likelihood of such hallucinations. Which approach should they take?

Accepted Answer

Provide the article text as context in the prompt and instruct the model to only use that information.. Hallucinations occur when a model generates plausible but unsupported information. To mitigate, you can ground the model by providing the source text in the prompt and instructing it to rely solely on that text. This constrains the model's output to the given context, reducing the chance of introducing external or fabricated details. This is a standard technique in generative AI for summarization.

Answer

Decrease the top-k parameter to limit the vocabulary used in the summary.

Answer

Increase the temperature parameter to encourage more creative outputs.

Answer

Use a smaller model with fewer parameters to limit the amount of information it can generate.

Question 10

A small marketing agency wants to quickly experiment with prompts for Gemini to draft social media posts, without writing code or provisioning infrastructure. They need a Google Cloud console experience for iterating on prompts and comparing model outputs. Which offering should they use?

Accepted Answer

Vertex AI Studio. Vertex AI Studio is the console-based workspace for prompt design, model comparison, and rapid prototyping with Gemini. It requires no coding or infrastructure setup, making it ideal for a marketing team that wants to experiment with social post drafts. The other choices are pipeline orchestration, SQL-based modeling, or office productivity tools that do not offer interactive prompt experimentation.

Answer

BigQuery ML

Answer

Gemini for Google Workspace

Answer

Vertex AI Pipelines

Question 11

A bank is piloting a Gemini-powered assistant that summarizes internal audit reports. Compliance requires that prompts and responses never leave the company's chosen Google Cloud region, that customer-managed encryption keys protect data at rest, and that no data is used to improve Google's models. Which combination of Vertex AI capabilities should the team configure to meet these requirements?

Accepted Answer

Vertex AI with data residency controls, CMEK through Cloud KMS, and the documented data governance guarantee that customer data is not used to train Google's foundation models.. Meeting all three compliance constraints requires regional processing for residency, Cloud KMS customer-managed keys for encryption at rest, and reliance on Vertex AI's data governance commitments that customer data is not used to train Google's foundation models. Network or prompt-level controls cannot substitute for these platform-level guarantees.

Answer

Vertex AI with CMEK enabled and a VPC Service Controls perimeter, relying on the perimeter to guarantee regional data processing.

Answer

Vertex AI with a global endpoint, default Google-managed encryption keys, and a prompt instruction telling the model not to retain any input.

Answer

Gemini accessed through AI Studio with billing enabled, plus a Cloud Armor policy restricting access to the bank's IP ranges.

Question 12

A team is designing prompts for a generative AI model to summarize legal documents. They want to improve the quality and relevance of the summaries. Which two prompt engineering best practices should they follow? (Choose two.)

Accepted Answer

Include a few examples of well-written summaries in the prompt.. Clear instructions and few-shot examples are proven prompt engineering techniques that enhance output quality. They reduce ambiguity and guide the model toward the desired format and content. Other options like high temperature or step-by-step reasoning are not appropriate for precise summarization tasks.

Answer

Keep the prompt as short as possible to avoid confusing the model.

Answer

Ask the model to think step by step before providing the summary.

Answer

Use the highest possible temperature setting to encourage creativity.

Question 13

A marketing team is using Google Cloud's Vertex AI Studio to generate product descriptions. They want the model to produce output in a very specific JSON format so it can be parsed by their downstream application. Which feature should they use to reliably constrain the model's output to that structure?

Accepted Answer

Controlled generation with a response schema. Constrained decoding through a response schema forces the model's tokens to conform to a predefined structure, so the output is guaranteed to be valid JSON with the expected fields. Prompt-level instructions and sampling parameters only influence content probabilistically and cannot guarantee parseable structure. For integration with downstream systems that require strict formatting, the schema-based approach is the only reliable choice.

Answer

Adding a system instruction that says 'output JSON'

Answer

Increasing the temperature parameter to 1.0

Answer

Setting the top-P parameter to 0

Question 14

A retail company is building a generative AI assistant on Google Cloud that drafts personalized product recommendations in natural language. Business leaders want to measure whether the pilot is delivering value before expanding it to all customers. They need a metric that reflects ongoing operational benefit rather than one-time build effort. Which metric should they prioritize?

Accepted Answer

The percentage of recommendations accepted by customers, tracked over time.. Tracking the share of generated recommendations that customers accept over time ties the generative AI assistant to a sustained business outcome rather than a one-time build activity. It is observable in production, supports comparison against baselines, and gives leadership a defensible signal for deciding whether to scale the pilot. Infrastructure and authoring counts describe effort, not delivered value, so they cannot justify expansion on their own.

Answer

The total GPU hours consumed while fine-tuning the recommendation model.

Answer

The number of prompt templates authored during the pilot.

Answer

The number of Vertex AI endpoints created for the pilot environment.

Question 15

A financial analytics firm wants to query its BigQuery data using natural language, without exporting data to a separate service. They need a solution that integrates directly with BigQuery and uses Gemini models to generate SQL and interpret results. Which Google Cloud capability should they use?

Accepted Answer

Gemini in BigQuery. Gemini in BigQuery embeds generative AI capabilities directly into BigQuery, allowing users to ask questions in plain language, automatically generate SQL, and get explanations of results. This eliminates the need to export data or build custom integrations, making it the ideal choice for a financial firm that wants to query BigQuery data conversationally.

Answer

Cloud SQL with Gemini

Answer

Vertex AI Search

Answer

BigQuery ML with Gemini models

Question 16

A developer is using Vertex AI's text generation model to create product descriptions. They want to ensure the output adheres to a specific brand voice and style. Which approach is most effective without retraining the model?

Accepted Answer

Use few-shot prompting with examples of brand-consistent descriptions.. Few-shot prompting is a powerful, no-retraining method to steer a model's style by providing examples. It allows the model to infer the desired tone and structure from the prompt. Other options either require retraining or adjust parameters that affect randomness, not style adherence.

Answer

Fine-tune the model on a dataset of brand-approved descriptions.

Answer

Increase the top-k parameter to allow more diverse word choices.

Answer

Set the temperature to a high value to encourage creativity.

Question 17

A healthcare provider uses a Gemini model on Vertex AI to summarize patient intake notes for clinicians. The summaries must consistently follow a fixed structure: chief complaint, history, medications, and plan. Early tests show the model sometimes reorders or omits sections. Which technique most reliably enforces the required structure?

Accepted Answer

Provide few-shot examples in the prompt that show the exact section order and format.. Few-shot examples are the most reliable prompt-level technique for enforcing a fixed output structure, because they show the model the exact sections and their order within the same request. They need no retraining and can be revised as the template changes. Temperature changes increase disorder, broad unstructured fine-tuning does not teach the template, and token limits can drop sections rather than order them.

Answer

Fine-tune the model on a large corpus of unstructured clinical literature.

Answer

Increase the temperature so the model varies its section ordering.

Answer

Reduce the maximum output tokens to force the model to be concise.

Question 18

A team is using a generative AI model to answer customer questions about a complex product. They want to improve the factual accuracy and reduce hallucinations. Which TWO techniques should they apply? (Choose two.)

Accepted Answer

Use a grounded prompt that instructs the model to answer only from provided context and to say 'I don't know' if unsure.. Grounding with RAG supplies authoritative context, and a grounded prompt instructs the model to use only that context and to express uncertainty when appropriate. Together they attack hallucinations at the source: the model no longer relies solely on memorized parameters and is discouraged from fabricating. Other techniques like temperature or top-k affect randomness, not factual grounding, and fine-tuning on generic data does not provide the needed product facts.

Answer

Fine-tune the model on a small set of generic customer service dialogues.

Answer

Reduce the top-k parameter to a very small value.

Answer

Increase the temperature to make the model more confident.

Question 19

A product team is new to generative AI and wants to understand which characteristics distinguish foundation models from traditional task-specific ML models. Which two statements accurately describe foundation models? (Choose two.)

Accepted Answer

They can generate novel content such as text, images, or code rather than only classifying existing inputs.. Foundation models are pretrained on broad data, adapt to many tasks through prompting or fine-tuning, and can generate new content across modalities. These traits explain their flexibility and rapid reuse. The other statements describe classical supervised learning, require impractical retraining, or incorrectly claim foundation models are small and cheap, which misrepresents their scale and cost profile.

Answer

They are always smaller and cheaper to run than task-specific models because they do less work.

Answer

They must be trained from scratch for every new use case before they can produce any useful output.

Answer

They typically require a large labeled dataset specific to each target task before they can generate text.

Question 20

A financial services company wants to build a generative AI application that drafts personalized emails to clients. They require the model to be accessible via a fully managed API with minimal infrastructure management. Which Google Cloud service should they use?

Accepted Answer

Vertex AI Gemini API. The Vertex AI Gemini API offers a managed, serverless way to access powerful generative models. It eliminates infrastructure management, supports text generation, and integrates with Google Cloud's security and monitoring. Other options are either not generative or require custom model training.

Answer

AutoML Natural Language

Answer

Dialogflow CX

Answer

Cloud Natural Language API