Generative AI Leader Practice Exam 3 — 20 Questions

Question 1

A company is deploying a generative AI system for resume screening. They want to ensure fairness and avoid bias. Which TWO actions should they take? (Choose 2)

Accepted Answer

Evaluate the model's decisions across gender and ethnicity groups using a diverse test set. Option A is correct because fairness auditing requires measuring model outcomes across demographic subgroups (e.g., gender, ethnicity) using a diverse, representative test set, so disparate impact or bias can be detected and quantified before deployment. Option B is correct because if the training data is not representative of the candidate population, the model will learn skewed patterns and produce biased predictions, so ensuring representative training data is a foundational bias-mitigation step. Option C is incorrect because SynthID is a watermarking technique for identifying AI-generated content, not a fairness or bias-mitigation control for screening decisions. Option D is incorrect because simply removing protected attributes does not eliminate bias, since proxy variables and correlated features can still encode discriminatory patterns. Option E is incorrect because a larger model improves capacity or accuracy but does not inherently ensure fairness or reduce bias.

Answer

Apply SynthID watermarking to all generated decisions

Answer

Remove protected attributes like gender and race from training data

Answer

Use a larger model to improve accuracy

Question 2

A financial services firm is developing a GenAI application for investment advice. They need to ensure regulatory compliance. Which business strategy should they prioritize?

Accepted Answer

Implement strict human-in-the-loop review for all investment recommendations. In regulated industries like financial services, GenAI applications must prioritize compliance over speed. Option B is correct because a human-in-the-loop (HITL) review ensures that every investment recommendation is auditable and meets regulatory standards (e.g., SEC or FINRA rules), mitigating risks of hallucinated or non-compliant outputs. This strategy directly addresses the need for accountability and transparency in high-stakes decision-making.

Answer

Rapidly deploy an MVP and iterate based on user feedback

Answer

Open-source the model to gain community trust

Answer

Partner with a cloud provider that offers indemnification for model outputs

Question 3

Which TWO features are available in Vertex AI Studio for prompt engineering? (Choose two.)

Accepted Answer

Side-by-side comparison of model outputs. Vertex AI Studio provides a side-by-side comparison feature that allows prompt engineers to evaluate outputs from multiple model configurations or parameter settings simultaneously. This enables direct visual comparison of responses, helping to identify the most effective prompt phrasing or parameter combination without manual switching.

Answer

One-click deployment to a Vertex AI endpoint

Answer

Fine-tuning models directly in the interface

Answer

Building conversational agents with drag-and-drop

Question 4

Which TWO features are available in Vertex AI Agent Builder to enhance the conversational abilities of an agent? (Choose TWO.)

Accepted Answer

Slot filling. Slot filling is correct because it allows the agent to collect required parameters (slots) from the user during a conversation, enabling multi-turn interactions to fulfill complex requests. In Vertex AI Agent Builder, slot filling is a core feature for conversational agents, as it systematically prompts for missing information (e.g., date, location) until all necessary slots are filled, enhancing the agent's ability to handle dynamic user inputs.

Answer

Sentiment analysis

Answer

Code execution

Answer

Knowledge base integration

Question 5

A company uses a generative model to produce product descriptions. The descriptions are factually inconsistent with the product specs. Which technique would best ensure factual accuracy?

Accepted Answer

Implement retrieval-augmented generation (RAG) with product database. Retrieval-augmented generation (RAG) is the best technique because it dynamically retrieves relevant, up-to-date product specifications from a trusted database at inference time, grounding the model's output in verified facts. This directly addresses factual inconsistency by ensuring the generated description is based on authoritative source data rather than relying solely on the model's parametric memory.

Answer

Enhance the system prompt with product details

Answer

Lower the temperature to 0.0

Answer

Fine-tune the model on product descriptions

Question 6

Which TWO Google Cloud services can be used together to implement a RAG (retrieval-augmented generation) pipeline? (Select 2)

Accepted Answer

Vertex AI Vector Search. Vertex AI Vector Search (option B) is correct because it provides a managed vector database for storing and querying embeddings, which is essential for the retrieval step in a RAG pipeline. It enables semantic similarity search over large datasets, allowing the system to fetch relevant context documents based on a user query.

Answer

Cloud SQL

Answer

Bigtable

Answer

Cloud Functions

Question 7

Refer to the exhibit. The team changed the generation parameters to reduce output variability. However, summaries now often repeat the same phrases. Which parameter change is most likely causing the repetition?

Accepted Answer

Reducing temperature from 0.7 to 0.2. Reducing temperature from 0.7 to 0.2 makes the model's token selection much more deterministic, favoring the highest-probability tokens. This low-entropy sampling often causes the model to repeat the same phrases across summaries because it consistently picks the most likely continuation.

Answer

Reducing top_p from 0.95 to 0.85

Answer

Using the same model text-bison@002

Answer

Reducing top_k from 40 to 10

Question 8

A startup wants to deploy a custom-tuned large language model for real-time inference on Vertex AI. They need the lowest possible latency for end users. What deployment strategy should they choose?

Accepted Answer

Deploy the tuned model to a Vertex AI endpoint with GPU acceleration and autoscaling.. Deploying the custom-tuned model to a Vertex AI endpoint with GPU acceleration and autoscaling is the best choice for lowest-latency real-time inference. A dedicated endpoint keeps the model loaded and ready to serve requests, avoiding the per-request startup overhead of serverless wrappers like Cloud Functions. GPU acceleration speeds up the model's forward-pass computation, and autoscaling adds or removes replicas to match traffic so that requests are not queued behind insufficient capacity. Batch prediction (D) is designed for high-throughput offline jobs and is not suitable for interactive, low-latency use cases.

Answer

Use Vertex AI Model Garden to deploy the base PaLM 2 model.

Answer

Wrap the model in a Cloud Function and invoke via HTTP.

Answer

Use Vertex AI Batch Prediction to process requests in batches.

Question 9

A company uses a text-to-image model to generate marketing visuals. The results often misinterpret the prompt, e.g., 'a red car' generates a blue car. Which technique should they try first to align the output with the prompt?

Accepted Answer

Refine the prompt with more adjectives and context, e.g., 'bright red sports car'. Refining the prompt with more adjectives and context directly addresses the root cause of misalignment: insufficient specificity in the text description. Text-to-image models rely on the semantic richness of the prompt to guide the latent diffusion process; adding 'bright red sports car' provides stronger conditioning signals that steer the model's cross-attention layers toward the intended color and object attributes. This is the most efficient first step before adjusting hyperparameters like guidance scale.

Answer

Use a negative prompt to exclude blue

Answer

Upscale the image resolution to 1024x1024

Answer

Increase the guidance scale to 20

Question 10

Which Google Cloud service provides a managed environment for prompt engineering and model evaluation?

Accepted Answer

Vertex AI Generative AI Studio. Vertex AI Generative AI Studio is the correct answer because it is a managed service within Vertex AI specifically designed for prompt engineering, model tuning, and evaluation of generative AI models. It provides a no-code interface for testing prompts, comparing model outputs, and iterating on prompt design, directly supporting the workflow described in the question.

Answer

AI Platform Notebooks

Answer

Dialogflow CX

Answer

Cloud Composer

Question 11

A tech company wants to ensure that their generative AI model does not produce harmful content. They plan to use Google Cloud's content safety features. Which two methods can they use to customize content safety? (Choose two.)

Accepted Answer

Adjust safety thresholds for different categories like hate speech and violence. Option C is correct because Google Cloud's generative AI safety features (such as those in Vertex AI) let you configure per-category safety thresholds, so you can raise or lower sensitivity for categories like hate speech, harassment, sexually explicit content, and violence to match your tolerance policy. Option D is correct because you can supply a custom blocklist of prohibited terms or phrases, which the service checks in addition to the built-in safety filters, giving you organization-specific control over banned content. Option A is not a customization method—using default filters unchanged is the absence of customization. Option B is wrong because disabling all safety filters removes protection rather than customizing it, and it is not a supported way to ensure harmful content is blocked. Option E is not part of Google Cloud's content safety customization features; training a separate detection model is a different, external approach rather than configuring the built-in safety controls.

Answer

Use the default safety filters without any modifications

Answer

Disable all safety filters for maximum model creativity

Answer

Train a separate model to detect harmful content

Question 12

A healthcare organization needs to run ML models on patient data stored in BigQuery while ensuring data never leaves the database. Which service allows them to create and execute ML models directly in BigQuery SQL?

Accepted Answer

BigQuery ML. BigQuery ML enables SQL-based ML model creation and execution directly on data in BigQuery, meeting the requirement without data movement.

Answer

Cloud Functions

Answer

Vertex AI Prediction

Answer

Vertex AI Workbench

Question 13

An organization is running a large-scale training job for a custom NLP model with a batch size of 2048 and sequence length of 512. They need to minimize training time while keeping costs predictable. Which Google Cloud hardware should they choose?

Accepted Answer

Cloud TPU v5e pods. Cloud TPU v5e pods are purpose-built for large-scale training of transformer-based NLP models, offering high-throughput matrix multiplication and efficient scaling across multiple chips. With a batch size of 2048 and sequence length of 512, TPU v5e pods deliver superior training speed and predictable pricing via reserved capacity, minimizing time-to-train compared to GPU alternatives.

Answer

Compute Engine with NVIDIA A100 GPUs

Answer

Edge TPU devices

Answer

Compute Engine with NVIDIA T4 GPUs

Question 14

A media company uses a generative AI model to create marketing images. They want to ensure that AI-generated images can be identified as synthetic. Which Google Cloud capability should they use?

Accepted Answer

SynthID. SynthID is Google DeepMind's watermarking technology that embeds imperceptible digital watermarks into AI-generated content, including images, and provides a detector to identify them as synthetic. It is purpose-built for provenance and identification of generative AI output, matching the media company's requirement.

Answer

Cloud Vision API

Answer

Data Loss Prevention (DLP) API

Answer

Vertex AI Model Registry

Question 15

A team is fine-tuning a model for a legal document summarization task. They need to ensure high accuracy and avoid hallucinations. Which TWO approaches should they combine? (Choose two.)

Accepted Answer

Use Retrieval-Augmented Generation to retrieve relevant legal texts. Retrieval-Augmented Generation (RAG) is correct because it grounds the model's output in retrieved, authoritative legal texts, directly reducing hallucination by providing factual context during generation. This is critical for legal summarization where accuracy is paramount, as RAG ensures the model references specific statutes or case law rather than relying solely on its parametric memory.

Answer

Increase temperature to 1.5 during inference

Answer

Implement early stopping during fine-tuning

Answer

Use character-level tokenization to improve spelling

Question 16

In the transformer architecture, what is the role of the attention mechanism?

Accepted Answer

It decides which parts of the input to focus on when generating each token. The attention mechanism in the Transformer architecture computes a weighted sum of all input token representations, allowing the model to dynamically focus on the most relevant parts of the input sequence when generating each output token. This is achieved through learned query, key, and value projections that produce attention scores, enabling the model to capture long-range dependencies and contextual relationships. Option B correctly identifies this core function of selectively attending to input elements during token generation.

Answer

It normalizes the output of each layer

Answer

It predicts the next token directly

Answer

It converts tokens into numerical vectors

Question 17

An employee wants to use GenAI to assist with writing formulas in Google Sheets. Which Gemini for Google Workspace feature should they use?

Accepted Answer

Formula assistance in Google Sheets. Gemini for Sheets provides formula assistance, helping users generate, explain, or debug formulas.

Answer

Help me write in Google Docs

Answer

Image generation in Google Slides

Answer

Smart Compose in Gmail

Question 18

A data scientist is using Vertex AI RAG Engine to build a question-answering system over a large corpus of technical manuals. Users report that answers are often verbose and include irrelevant details. Which configuration change is MOST likely to improve answer conciseness?

Accepted Answer

Decrease the number of retrieved documents (k) in the retrieval step. In a RAG pipeline, the number of retrieved documents (k) directly controls how much context is fed to the LLM. Reducing k limits the amount of potentially irrelevant material the model sees, which typically makes answers shorter and more focused. This is the most direct lever for conciseness.

Answer

Increase the chunk size of documents in the index

Answer

Use a different embedding model for retrieval

Answer

Increase the max output token limit

Question 19

You are an ML engineer at a retail company. You have deployed a generative AI model on Vertex AI to generate product descriptions. The model uses a custom container and is deployed to a single endpoint. Recently, you noticed that inference latency has increased significantly during peak hours, causing timeouts. You have checked the logs and found that the CPU utilization on the deployed instances is consistently above 90% during peak hours. The model is currently deployed with a single machine type (n1-standard-4) and no scaling. You need to reduce latency without incurring excessive cost. What should you do?

Accepted Answer

Change the machine type to n1-standard-8 and enable autoscaling with min replicas=1, max replicas=5. Upgrading to a larger machine (n1-standard-8) provides more CPU cores to handle the increased inference workload, while enabling autoscaling (min=1, max=5) allows the deployment to dynamically add replicas during peak hours to distribute the load and reduce latency. This combination addresses the high CPU utilization without over-provisioning during off-peak times, thus controlling cost.

Answer

Optimize the model using quantization and reduce the number of replicas

Answer

Switch to batch prediction instead of online prediction

Answer

Add a GPU accelerator to the existing machine

Question 20

After deploying a text-to-image model, the output images often contain distorted objects. The team suspects the prompt is too complex. Which prompt engineering technique should they try first?

Accepted Answer

Break the prompt into simpler, separate steps.. Breaking a complex prompt into simpler, separate steps reduces the cognitive load on the diffusion model, allowing it to focus on generating each element sequentially. This technique, often called 'prompt decomposition' or 'step-by-step prompting,' directly addresses the root cause of distorted objects when the model struggles to attend to multiple conflicting details simultaneously in a single pass.

Answer

Increase the guidance scale.

Answer

Add more descriptive adjectives.

Answer

Use a negative prompt to exclude distortions.