Generative AI Leader Practice Test 41 — 15 Questions

Question 1

A company deployed a large language model on Vertex AI using the configuration shown in the exhibit. During peak usage, users report high latency. Which change is most likely to improve latency?

Accepted Answer

Increase minReplicaCount to 3.. Increasing minReplicaCount to 3 ensures that at least three instances of the model are always running and ready to serve requests. This reduces cold-start latency and distributes the load across multiple replicas, directly addressing high latency during peak usage by providing more concurrent serving capacity.

Answer

Remove the accelerator to simplify deployment.

Answer

Switch to a GPU with more memory, such as NVIDIA_TESLA_A100.

Answer

Change machineType to n1-standard-4 to reduce cost.

Question 2

Refer to the exhibit. A data scientist runs this command to upload a custom model to Vertex AI. What is the primary purpose of the --container-image-uri flag?

Accepted Answer

To specify the base image for model serving. The --container-image-uri flag in the `gcloud ai models upload` command specifies the custom container image that Vertex AI will use to serve predictions. This is the base image for model serving, not for training, because Vertex AI uses this image to create the serving environment that hosts the model and handles prediction requests.

Answer

To indicate the model artifact location

Answer

To set the training container

Answer

To define the prediction container

Question 3

A financial services firm needs to generate synthetic data for training models while ensuring that no real customer data leaks. Which technique should they use?

Accepted Answer

Differential privacy during fine-tuning. Differential privacy during fine-tuning is the correct technique because it adds calibrated noise to the training process, ensuring that the synthetic data generated does not reveal information about any individual real customer record. This approach provides a formal mathematical guarantee of privacy, making it suitable for generating synthetic data that preserves statistical properties while preventing data leakage. In contrast, other methods like redaction, masking, or using a public model do not inherently prevent the model from memorizing and reproducing sensitive information.

Answer

Using the Vertex AI PII redaction service

Answer

Using a public foundation model without fine-tuning

Answer

Data masking before training

Question 4

A company wants to build an internal knowledge base that allows employees to ask questions about company policies in natural language. The knowledge base is stored in a Google Cloud SQL database. Which architecture should they use?

Accepted Answer

Use Vertex AI Agent Builder with Grounding to connect to Cloud SQL. Vertex AI Agent Builder with Grounding allows connecting to the database and answering questions based on its content, providing a natural language interface.

Answer

Use Gemini API with a prompt that includes all policies

Answer

Use AutoML Natural Language to classify questions

Answer

Export Cloud SQL to BigQuery and use BigQuery ML

Question 5

A company needs to deploy a chatbot on a mobile device that must work offline. They want to use Gemini for natural language understanding but need minimal latency and no cloud dependency. Which Gemini model variant is most appropriate?

Accepted Answer

Gemini Nano. Gemini Nano is the most appropriate variant because it is specifically designed for on-device deployment, enabling offline operation with minimal latency. It is optimized for mobile devices through quantization and efficient architecture, allowing natural language understanding without any cloud dependency.

Answer

Gemini Flash

Answer

Gemini Pro

Answer

Gemini Ultra

Question 6

A generative AI model for chatbot responses sometimes produces toxic language. The team wants to reduce toxicity without significantly affecting the model's helpfulness. Which approach is best?

Accepted Answer

Fine-tune with a dataset of non-toxic responses and use RLHF. Fine-tuning with a curated dataset of non-toxic responses directly adjusts the model's weights to reduce the likelihood of generating toxic language, while RLHF (Reinforcement Learning from Human Feedback) further aligns the model with human preferences for helpfulness and safety. This combined approach addresses the root cause of toxicity in the model's behavior without the blunt trade-offs of other methods, preserving the model's utility.

Answer

Increase the temperature parameter

Answer

Reduce the maximum output tokens

Answer

Apply a toxicity classifier as a post-processing filter

Question 7

A large enterprise runs a production application that uses the Gemini API on Vertex AI for real-time content moderation. They are experiencing occasional 429 (Too Many Requests) errors during peak hours. Their current quota is 1000 requests per minute (RPM) and they are hitting around 950 RPM on average, with spikes up to 1050. They have already implemented exponential backoff and retry logic. They need to reduce the error rate without reducing the quality of moderation. Which additional measure should they take?

Accepted Answer

Implement a local caching layer for common moderation queries.. Implementing a local caching layer for common moderation queries reduces the number of identical requests sent to the Gemini API, directly lowering the effective RPM without compromising moderation quality. Since the enterprise is already using exponential backoff and retry logic, caching addresses the root cause of hitting quota limits by eliminating redundant API calls, which is a standard pattern for rate-limit mitigation in production AI workloads.

Answer

Deploy the model on a dedicated Vertex AI endpoint with autoscaling.

Answer

Switch to a lower-tier model like Gemini 1.0 Pro to reduce quota consumption.

Answer

Request a quota increase from Google Cloud support.

Question 8

A startup with limited budget wants to quickly test a generative AI use case for personalized email marketing. Which approach minimizes time-to-market and cost?

Accepted Answer

Use a managed API like the PaLM API with prompt engineering.. Using a managed API like the PaLM API with prompt engineering eliminates the need for infrastructure setup, model training, and data preparation. This approach leverages a pre-trained model via a simple REST API call, allowing the startup to iterate on prompts and achieve personalized email content in hours rather than weeks, minimizing both time-to-market and cost.

Answer

Hire a team of AI researchers to build a solution.

Answer

Develop a custom model from scratch.

Answer

Fine-tune a large open-source model on internal data.

Question 9

A healthcare company wants to use generative AI to summarize patient records but must comply with HIPAA. Which deployment option should they choose?

Accepted Answer

Use an on-premises deployment of open-source model. An on-premises deployment of an open-source model ensures that all patient data remains within the organization's controlled infrastructure, never leaving the local network. This eliminates any risk of data transmission to external cloud services, which is critical for HIPAA compliance where protected health information (PHI) must be safeguarded against unauthorized access or breaches. On-premises solutions allow the organization to implement its own security controls, encryption, and audit trails without relying on a third-party's compliance posture.

Answer

Use Vertex AI on Google Cloud with data residency

Answer

Use Google Workspace AI

Answer

Use a third-party API

Question 10

A gen AI application produces hallucinations (factually incorrect outputs). Which mitigation strategy is LEAST effective?

Accepted Answer

Increasing model temperature. Increasing model temperature makes the model more random and creative, which directly increases the likelihood of hallucinations. It does not constrain or ground the output in factual data, making it the least effective mitigation strategy among the options.

Answer

Using prompt templates with constraints

Answer

Using grounding with a knowledge base

Answer

Implementing retrieval-augmented generation

Question 11

A machine learning engineer is deploying a large generative model on Vertex AI. The model requires a GPU with high memory. Which machine configuration should they choose?

Accepted Answer

a2-highgpu-4g with 4 A100 GPUs. The a2-highgpu-4g machine series is specifically designed for large-scale GPU-accelerated workloads, offering 4 NVIDIA A100 GPUs with 40GB of high-bandwidth memory (HBM2e) each, totaling 160GB of GPU memory. This configuration provides the high memory capacity required for training or serving large generative models, such as LLMs or diffusion models, which often exceed the memory limits of smaller GPUs.

Answer

c2-standard-16 with no GPU

Answer

n1-standard-4 with a single T4 GPU

Answer

n2-standard-8 with a single P4 GPU

Question 12

A company is using Vertex AI Gemini API to analyze customer feedback. They notice that the model occasionally generates offensive content. They have already set safety settings to block high-probability harmful content. What additional step should they take to further reduce offensive outputs?

Accepted Answer

Adjust safety settings to block medium-probability harmful content. The company has already blocked high-probability harmful content, but offensive outputs can still occur at lower probability thresholds. By adjusting safety settings to block medium-probability harmful content, they tighten the filter to catch more borderline cases without requiring model retraining or sacrificing output diversity. This leverages Vertex AI's configurable safety filters, which operate on likelihood categories (e.g., high, medium, low) rather than just binary blocking.

Answer

Set the temperature to 0.0

Answer

Enable context caching

Answer

Fine-tune the model on customer feedback data

Question 13

A team is developing a mobile app that must run AI inference on-device for low latency and offline capability. Which Gemini model variant is designed specifically for on-device deployment?

Accepted Answer

Gemini Nano. Gemini Nano is the smallest and most efficient model in the Gemini family, specifically optimized for on-device deployment. It is designed to run directly on mobile devices (e.g., Android phones) using hardware acceleration like Google's Pixel Neural Core or Qualcomm's AI Engine, enabling low-latency inference and offline capability without requiring a cloud connection.

Answer

Gemini Pro

Answer

Gemini Ultra

Answer

Gemini Flash

Question 14

A company is using Vertex AI to generate email responses. They want to ensure sensitive customer data (PII) is not included in the output. What is the most effective approach?

Accepted Answer

Configure safety filters to block PII categories.. Safety filters in Vertex AI are specifically designed to block categories of harmful content, including PII, at the model's output layer. This provides a deterministic, automated guardrail that prevents sensitive data from being generated, unlike prompt-based instructions which can be overridden by the model's training. Safety filters operate on the model's response before it is returned, ensuring PII is caught even if the model attempts to generate it.

Answer

Use a system prompt instructing the model to avoid PII.

Answer

Fine-tune the model on a dataset that excludes PII.

Answer

Manually review each output before sending.

Question 15

Refer to the exhibit. This JSON describes a Vertex AI endpoint with a deployed model. Which statement about scaling is true?

Accepted Answer

The endpoint uses only dedicated resources, no automatic scaling. The JSON shows that the endpoint is configured with `dedicatedResources` and no `autoscalingMetricSpecs` or `minReplicaCount`/`maxReplicaCount` fields. In Vertex AI, when you specify only `machineSpec` and a fixed `minReplicaCount` (here implicitly 1) without a `maxReplicaCount` or autoscaling metrics, the endpoint uses dedicated resources with no automatic scaling — the model will always run on exactly the number of replicas you define, regardless of load.

Answer

The endpoint will automatically scale based on GPU utilization

Answer

The endpoint will scale from 1 to 3 replicas based on load using automatic scaling

Answer

The endpoint can scale to zero when not in use