Generative AI Leader Fundamentals of Generative AI — 100 Questions

Question 1

A data scientist notices that a text generation model deployed on Vertex AI returns repetitive outputs after a few turns in a chat application. What is the most likely cause and the best parameter adjustment?

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

Accepted Answer

The top_p value is too high; reduce top_p to limit token sampling.. Repetitive outputs in a chat application after a few turns are typically caused by the model getting stuck in a loop due to high cumulative probability from top-p sampling. Reducing top_p limits the set of tokens considered at each step, forcing the model to explore less likely tokens and breaking the repetition cycle. This directly addresses the issue without sacrificing coherence, unlike temperature adjustments which affect randomness globally.

Answer

The max_output_tokens is too low; increase it to allow more diverse output.

Answer

The model is overfitted; switch to a smaller model.

Answer

The temperature is too low; increase temperature to add randomness.

Question 2

A financial services company wants to use generative AI to generate personalized investment advice. They must ensure responses comply with regulatory requirements (e.g., no guarantees of returns). Which Vertex AI safety feature should they primarily use?

Accepted Answer

Safety filters with a custom blocklist that includes phrases like 'guaranteed return'.. Safety filters with a custom blocklist allow the company to define specific prohibited phrases (e.g., 'guaranteed return') that the model must avoid generating. This provides a deterministic, rule-based enforcement layer that directly addresses regulatory compliance by blocking disallowed content at inference time, without relying on the model's probabilistic behavior.

Answer

Vertex AI Grounding with their compliance database.

Answer

Prompt engineering with instructions to avoid guarantees.

Answer

Reinforcement learning from human feedback (RLHF) on the model.

Question 3

A company is using Vertex AI to generate marketing copy. They notice that the output sometimes contains factual inaccuracies. Which parameter adjustment is most likely to improve factual accuracy?

Accepted Answer

Decrease the temperature parameter.. Decreasing the temperature parameter reduces the randomness of the model's output, making it more deterministic and less likely to generate creative but factually incorrect content. Lower temperature (e.g., 0.1) forces the model to choose higher-probability tokens, which aligns with more factual and consistent responses, especially in tasks like marketing copy where accuracy is critical.

Answer

Increase the max_output_tokens parameter.

Answer

Increase the top_p parameter.

Answer

Add a post-processing step to verify facts using a database.

Question 4

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 5

A team is evaluating generative AI models for a content moderation system. Which THREE metrics are most important to assess?

Accepted Answer

Percentage of outputs flagged by safety filters.. Option A is correct because the percentage of outputs flagged by safety filters directly measures how often the generative model produces content that violates safety policies, which is central to a content moderation system's purpose. Option C is correct because inference latency under expected load determines whether the moderation system can process requests in real time at production scale, a critical operational requirement. Option E is correct because precision and recall on a labeled test set of moderated content quantify the model's ability to correctly identify violating versus non-violating content, the core classification performance for moderation. Option B is not selected because cost per million tokens is a general cost-efficiency metric, not a primary measure of moderation quality or safety. Option D is not selected because BLEU score measures n-gram overlap with reference text and is unsuitable for evaluating moderation decisions against guidelines, which are classification criteria rather than text to be reproduced.

Answer

Cost per million tokens.

Answer

BLEU score against human-written moderation guidelines.

Question 6

A company wants to build a chatbot using Vertex AI that can answer customer questions based on their internal knowledge base. Which Google Cloud service should they use to store and retrieve the knowledge base efficiently?

Accepted Answer

Vertex AI Vector Search. Vertex AI Vector Search is the correct choice because it is purpose-built for semantic similarity search over embeddings, enabling the chatbot to retrieve relevant chunks from the knowledge base based on meaning rather than exact keyword matches. It integrates natively with Vertex AI and supports high-dimensional vector indexing, making it efficient for large-scale retrieval-augmented generation (RAG) workflows.

Answer

Cloud Storage

Answer

BigQuery

Answer

Vertex AI Matching Engine

Question 7

A company is using Vertex AI to generate personalized marketing emails. The model sometimes produces biased content. What is the most effective way to detect and mitigate bias?

Accepted Answer

Use Vertex AI Explainable AI to analyze predictions and detect bias in training data. Vertex AI Explainable AI provides feature attributions and model explanations that help identify which input features (e.g., demographic attributes, phrasing patterns) contribute most to biased outputs. By analyzing these attributions against training data, you can pinpoint and mitigate bias at the source, rather than relying on post-hoc manual review or model swapping.

Answer

Add more diverse training data

Answer

Manually review all generated emails before sending

Answer

Switch to a different generative model

Question 8

Which THREE factors should be considered when choosing between fine-tuning and prompt engineering for a generative AI task? (Choose three.)

Accepted Answer

Availability of labeled training data. Option A (Availability of labeled training data) is correct because fine-tuning requires a substantial, high-quality labeled dataset for the target task, whereas prompt engineering needs little or no labeled data, so data availability directly drives the choice. Option C (Latency requirements for the application) is correct because fine-tuned smaller models can be served with lower inference latency than large general-purpose models invoked with long, elaborate prompts, making latency a key decision factor. Option D (Degree of task specialization required) is correct because highly specialized or domain-specific tasks that demand consistent, format-constrained outputs often benefit from fine-tuning, while general tasks are well served by prompt engineering. Option B (Cost of API calls per request) is not one of the three factors because it is a general operational cost consideration rather than a defining criterion between the two approaches. Option E (Size of the base model) is not a deciding factor because both fine-tuning and prompt engineering can be applied to models of various sizes; model size alone does not determine which technique to use.

Answer

Cost of API calls per request

Answer

Size of the base model

Question 9

A data scientist sees the above error when trying to deploy a model to an endpoint. What is the most likely cause?

Accepted Answer

The model import into Vertex AI Model Registry is still in progress. The error indicates that the model is not yet fully imported into the Vertex AI Model Registry. Deploying a model to an endpoint requires the model resource to be in an 'ACTIVE' state; if the import is still in progress, the deployment request will fail. This is a common timing issue when a model is uploaded but not yet registered.

Answer

The IAM permissions are insufficient

Answer

The endpoint does not exist

Answer

The model is already deployed to another endpoint

Question 10

A developer receives the above JSON response from a Vertex AI PaLM API call for a medical advice application. What should the developer be most concerned about?

Accepted Answer

The output falls under the 'health' category, which may require compliance with regulations. The JSON response includes a 'category' field with the value 'health', which triggers stringent regulatory compliance requirements such as HIPAA in the US or GDPR in Europe. For a medical advice application, the developer must ensure data handling, model transparency, and output validation meet these legal standards, as failure could result in severe penalties. The PaLM API's safety attributes and category labels are designed to flag such sensitive domains, making compliance the primary concern over other technical issues.

Answer

The safety score is very low (0.01)

Answer

The deployed model ID is not recognized

Answer

The prediction content is incorrect

Question 11

An organization uses a fine-tuned model for medical diagnosis and must comply with HIPAA. Which measure is essential when deploying the model on Vertex AI?

Accepted Answer

Use a private Google Cloud Access and disable external internet access for the endpoint.. HIPAA requires that patient data be protected from unauthorized access during transmission and deployment. Using a private Google Cloud Access endpoint with external internet access disabled ensures that the model endpoint is only reachable within the organization's VPC network, preventing data exposure over the public internet and meeting HIPAA's security rule for safeguarding electronic protected health information (ePHI).

Answer

Store all patient data in Cloud Storage with object versioning.

Answer

Enable encryption at rest for all resources.

Answer

Use a publicly accessible endpoint for faster response times.

Question 12

A team is training a custom foundation model using JAX on TPUs on Google Cloud. They encounter frequent Out of Memory (OOM) errors. Which action is most effective in resolving the OOM error?

Accepted Answer

Enable model parallelism using GSPMD to distribute the model across TPU cores.. OOM errors when training large foundation models on TPUs often stem from the model exceeding the memory of a single TPU core. GSPMD (Generalized SPMD) enables automatic model parallelism, sharding the model's parameters, gradients, and optimizer states across multiple TPU cores, thereby reducing per-core memory pressure without altering the model architecture or precision.

Answer

Reduce the model size by decreasing the number of layers.

Answer

Increase the batch size to maximize TPU utilization.

Answer

Use mixed precision training (bfloat16) to reduce memory footprint.

Question 13

A data scientist is selecting a base model for generating Python code. Which TWO factors are most important to consider?

Accepted Answer

Model's license (proprietary vs open-source).. Option A is correct because the model's license determines whether the base model can legally be used, modified, and deployed in the data scientist's intended product or workflow, which is a fundamental gating factor when selecting a base model. Option B is correct because performance on coding benchmarks such as HumanEval directly measures a model's ability to generate correct Python code, making it the most relevant indicator of task-specific capability for code generation. Option C is not essential here because the scenario specifically targets Python code generation, so broad multi-language support is not a primary requirement. Option D is not a top factor because training data recency is a secondary consideration that does not directly guarantee code-generation quality. Option E is not decisive because parameter count alone does not reliably predict coding performance; a smaller model can outperform a larger one on HumanEval depending on training and architecture.

Answer

Model's support for multiple programming languages.

Answer

Model's training data recency.

Answer

Model's parameter count (size).

Question 14

A company deploys a Gemini model on Vertex AI for a customer-facing chatbot. They observe the chatbot occasionally produces toxic language. Which TWO measures should they implement immediately to reduce toxic outputs?

Accepted Answer

Enable the 'block offensive content' flag in the model's safety configuration.. Option D is correct because Vertex AI's safety configuration supports a 'block offensive content' flag (via safetySettings with harm categories such as HARM_CATEGORY_HATE_SPEECH, HARM_CATEGORY_HARASSMENT, HARM_CATEGORY_SEXUALLY_EXPLICIT, and HARM_CATEGORY_DANGEROUS_CONTENT) that filters toxic responses before they reach users. Option E is correct because safety thresholds (BLOCK_LOW_AND_ABOVE, BLOCK_MEDIUM_AND_ABOVE, BLOCK_ONLY_HIGH, BLOCK_NONE) can be configured on the Vertex AI endpoint/deployment to block hate speech and toxic content at inference time, providing an immediate guardrail. Option A is wrong because raising temperature increases randomness and creativity, making outputs less conservative, not more. Option B is wrong because adding another LLM to rephrase outputs is not an immediate, deterministic safety control and can itself introduce toxicity or latency. Option C is wrong because fine-tuning on polite conversations is a longer-term remediation requiring data curation and retraining, not an immediate measure.

Answer

Increase the model's temperature to make outputs more conservative.

Answer

Use a separate language model to rephrase the outputs before sending to users.

Answer

Fine-tune the model on a curated dataset of polite conversations.

Question 15

Refer to the exhibit. This IAM policy is applied to a Vertex AI project. A user 'test@example.com' reports they cannot create a ModelEvaluationPipelineJob. Which action should the administrator take?

Accepted Answer

Add the user to the roles/aiplatform.admin role at the project level.. Roles/aiplatform.user does not have the permissions to create pipeline jobs; it only allows viewing and using models and endpoints. Roles/aiplatform.admin has full control, so adding the user to this role is the simplest fix. There is no roles/aiplatform.specialist; removing the service account would not help; and granting at the model level is insufficient for creating pipeline jobs.

Answer

Grant the user roles/aiplatform.specialist at the project level.

Answer

Add the user roles/aiplatform.user at the model level to allow pipeline creation.

Answer

Remove the service account from roles/aiplatform.admin.

Question 16

Refer to the exhibit. A user with this IAM role tries to deploy a model to a Vertex AI Endpoint but fails. What is the most likely reason?

Accepted Answer

The user needs the roles/aiplatform.deployer role. The user has an IAM role but lacks the specific permission `aiplatform.deployments.create` required to deploy a model to a Vertex AI Endpoint. The `roles/aiplatform.deployer` role includes this permission, while the user's existing role does not, causing the deployment to fail. Even if the user can use other Vertex AI services, deploying a model to an endpoint is a distinct action that requires this specific role.

Answer

The user is not authorized to use Vertex AI at all

Answer

The model artifact is not in the same region as the endpoint

Answer

The user needs the roles/aiplatform.admin role

Question 17

A company has a large dataset of proprietary documents and wants to build a Q&A system using a foundation model without exposing the documents to the model. Which approach is most appropriate?

Accepted Answer

Use RAG with Vertex AI Vector Search and embeddings. Retrieval-Augmented Generation (RAG) with Vertex AI Vector Search allows the system to retrieve relevant document chunks via embeddings without exposing the full documents to the foundation model. The model only sees the retrieved context in the prompt, ensuring proprietary data remains isolated and not used for training or memorization.

Answer

Use a zero-shot model with context in prompt

Answer

Fine-tune the model on the documents

Answer

Use prompt engineering to instruct the model

Question 18

After fine-tuning a foundation model on company emails, the model outputs confidential information. What is the most likely cause?

Accepted Answer

The fine-tuning dataset was not anonymized. The most likely cause of a fine-tuned model outputting confidential information is that the fine-tuning dataset contained sensitive data that was not anonymized. During fine-tuning, the model learns patterns and can memorize specific sequences, including confidential details like names, addresses, or proprietary information, which it then reproduces in responses. This is a well-known data leakage risk in fine-tuning workflows.

Answer

The prompt is too vague

Answer

The model is too large

Answer

Overfitting to the training data leading to memorization

Question 19

A team is evaluating generative AI models on Vertex AI. They need to compare models based on specific criteria. Which TWO criteria are most important for selecting a model for a text summarization task?

Accepted Answer

ROUGE scores. For a text summarization task, ROUGE scores (option A) are the standard evaluation metric that directly measures n-gram overlap between generated summaries and reference summaries, making them the most relevant quality criterion for comparing models. Latency (option E) is also critical because summarization is typically deployed in interactive or near-real-time pipelines, so response time directly affects user experience and system throughput. Training dataset size (option B) is not a reliable selection criterion since it is often undisclosed and does not necessarily correlate with summarization quality. Cost per token (option C) is a deployment consideration but not a primary model-selection criterion for task quality, and model size in parameters (option D) is an architectural detail that does not directly indicate summarization performance.

Answer

Training dataset size

Answer

Cost per token

Answer

Model size in parameters

Question 20

Refer to the exhibit. A developer creates a model resource with this YAML config but gets an error that the model is not deployable. What is missing?

Accepted Answer

container_spec. The error 'model is not deployable' occurs because the YAML config lacks a `container_spec` field. In Vertex AI, a model must specify how to serve predictions—either via a pre-built container (using `container_spec`) or a custom container. Without this, the model has no runtime environment and cannot be deployed to an endpoint.

Answer

model_type

Answer

artifact_uri

Answer

description