Generative AI Leader Practice Test 50 — 15 Questions

Question 1

During model evaluation, a team observes good performance on training data but poor on validation data. Which regularization technique is most appropriate to address this?

Accepted Answer

Apply dropout. The scenario describes overfitting, where the model memorizes training data but fails to generalize to unseen validation data. Dropout is a regularization technique that randomly deactivates a fraction of neurons during training, forcing the network to learn more robust features and reducing co-adaptation, which directly mitigates overfitting.

Answer

Add more training data

Answer

Increase the learning rate

Answer

Use a larger batch size

Question 2

Which TWO of the following are best practices for prompt engineering?

Accepted Answer

Provide context and examples in the prompt. Option A is correct because supplying relevant context and few-shot examples in the prompt grounds the model, clarifies the expected format and intent, and measurably improves output relevance and accuracy. Option C is correct because clear, specific instructions reduce ambiguity, constrain the model's response space, and yield more predictable, on-target results. Together, these two practices are widely recommended in prompt engineering guidance for both large language models and generative AI services. Option B does not belong because adding random noise degrades coherence and reliability rather than improving genuine creativity. Option D does not belong because maximizing token count wastes context window and cost without improving quality; concise, purposeful prompts are preferred. Option E does not belong because negative prompts are a feature specific to certain image-generation tools and are not a general best practice for prompt engineering across LLMs.

Answer

Append random noise to prompts to improve creativity

Answer

Always use the maximum possible number of tokens

Answer

Use negative prompts to discourage undesired outputs

Question 3

A company is comparing Google Cloud Vertex AI with AWS Bedrock and Azure OpenAI. They need a model that can natively process text, images, audio, and video. Which differentiator does Google Cloud offer?

Accepted Answer

Gemini's native multimodal capabilities. Gemini is Google's multimodal model natively trained on text, images, audio, and video from the ground up, enabling it to process and reason across these modalities without separate components. This native capability is a key differentiator for Google Cloud Vertex AI, as competing platforms like AWS Bedrock and Azure OpenAI primarily offer models that are text-centric or require separate models for different modalities.

Answer

Access to GPT-4o

Answer

Support for Anthropic Claude

Answer

Integration with Microsoft Copilot

Question 4

What is the primary advantage of using embeddings and vector search for semantic search over traditional keyword search?

Accepted Answer

Ability to find documents with similar meaning even without exact keyword matches. Embeddings and vector search capture semantic meaning by converting text into high-dimensional vectors, enabling retrieval of documents with similar meaning even when they lack exact keyword matches. This is the primary advantage over traditional keyword search, which relies on literal term matching and fails with synonyms, paraphrases, or conceptual similarity.

Answer

Faster retrieval speed

Answer

Lower storage requirements

Answer

No need for indexing

Question 5

A data scientist fine-tunes a foundation model on customer support transcripts. After evaluation, the model's responses are too formal. Which adjustment during fine-tuning is most likely to make responses more conversational?

Accepted Answer

Include examples of informal customer interactions in the fine-tuning data.. The training data directly influences the tone and style of model outputs. Including examples of informal conversations in the fine-tuning dataset teaches the model the desired conversational tone. Other options affect training dynamics but not the style.

Answer

Increase the batch size to stabilize training.

Answer

Decrease the number of fine-tuning steps to prevent overfitting.

Answer

Use a higher learning rate for faster adaptation.

Question 6

Which Google Cloud tool allows you to document the intended use, training data, and evaluation results of a machine learning model to promote transparency?

Accepted Answer

Model Cards. Model Cards are designed to provide transparent documentation of ML models, including intended use, training data, and evaluation results.

Answer

PAIR Explorables

Answer

Datasheets for Datasets

Answer

People + AI Guidebook

Question 7

A data scientist is fine-tuning a generative AI model for customer sentiment analysis. To ensure the fine-tuned model does not inadvertently memorize and reproduce personally identifiable information (PII) from the training data, which THREE practices should they follow? (Select 3)

Accepted Answer

Apply differential privacy during fine-tuning. Option A is correct because applying differential privacy during fine-tuning adds calibrated noise to the training process, mathematically bounding the influence any single training record can have on the model's parameters and thereby limiting memorization of PII. Option C is correct because removing or anonymizing PII in the training data eliminates the sensitive content before it can ever be learned, which is the most direct defense against the model reproducing it. Option D is correct because data minimization restricts training to only the data necessary for the sentiment-analysis task, reducing the volume of PII exposed to the model and shrinking the attack surface for memorization. Option B is not correct because quantization only compresses weights to lower precision to reduce size and inference cost; it does not prevent the model from having memorized PII during training. Option E is not correct because k-fold cross-validation is an evaluation/resampling technique for estimating generalization performance and does not remove, mask, or otherwise protect PII in the training data.

Answer

Apply model quantization to reduce model size

Answer

Use k-fold cross-validation to evaluate the model

Question 8

A machine learning engineer is using a generative AI model that occasionally outputs biased content. Which TWO actions should they take to address this issue?

Accepted Answer

Adjust the training data to be more representative of the target population. Option C is correct because bias in generative AI often originates from unrepresentative or skewed training data, so adjusting the training data to better reflect the target population directly reduces the model's tendency to produce biased content. Option E is correct because evaluating model outputs on diverse test sets is essential for detecting bias across different demographic groups, since a single homogeneous test set can mask disparities in model performance. Options A, B, and D are not appropriate: using a single test set (A) limits bias detection and hides subgroup disparities, removing safety filters (B) would increase harmful or biased outputs rather than mitigate them, and increasing model size (D) does not inherently fix bias and may even amplify it if the underlying data remains unrepresentative.

Answer

Use a single test set for simplicity

Answer

Remove all safety filters to allow more freedom in outputs

Answer

Increase the model's size to improve generalization

Question 9

A company wants to use GenAI to create a knowledge base assistant that can answer questions from internal documentation. They need the assistant to always have access to the latest documents without retraining. Which TWO services should they combine? (Choose 2)

Accepted Answer

Vertex AI Agent Builder. Vertex AI Agent Builder (B) is correct because it provides the orchestration and grounding layer for building conversational search and RAG-style assistants over enterprise content, letting the agent retrieve from a connected data store at query time rather than relying on model retraining. Vertex AI Vector Search (E) is correct because it supplies the scalable nearest-neighbor index that stores embeddings of the internal documents and returns the most relevant chunks for each question, so newly added or updated documents can be indexed and served immediately. Together they satisfy the requirement that the assistant always accesses the latest documents without retraining, since retrieval happens dynamically against the vector index. Vertex AI Pipelines (A) is for orchestrating ML workflows such as training, tuning, and batch processing, not for serving up-to-date document retrieval to an assistant. Vertex AI Model Garden (C) is a catalog for discovering and deploying foundation models, which does not by itself provide document indexing or retrieval. Duet AI in Google Docs (D) is an end-user writing assistant inside Docs and cannot serve as the retrieval backend for a custom knowledge base assistant.

Answer

Vertex AI Pipelines

Answer

Vertex AI Model Garden

Answer

Duet AI in Google Docs

Question 10

Which TWO techniques are most effective for improving factual accuracy in a generative AI model's responses? (Choose two.)

Accepted Answer

Retrieval-Augmented Generation (RAG) with curated datasets.. Option A, Retrieval-Augmented Generation (RAG) with curated datasets, is correct because RAG retrieves relevant, up-to-date passages from an external curated corpus at inference time and injects them into the context, so the model's answer is anchored to verifiable source content rather than relying solely on parametric memory, which directly reduces hallucination and improves factual accuracy. Option C, grounding with a trusted knowledge base, is correct because grounding constrains generation to authoritative, domain-validated sources (e.g., enterprise knowledge bases or vetted document stores), letting the model cite or align with established facts and thereby improving correctness and traceability. Option B is wrong because raising temperature to 1.5 increases sampling randomness and diversity, which makes outputs less deterministic and more prone to fabrication, not more factual. Option D is wrong because longer system prompts with many instructions can dilute attention, introduce conflicting directives, and do not supply new factual evidence, so they are unreliable for accuracy. Option E is wrong because fine-tuning on a large general-text corpus mainly adapts style and broad language patterns and can even reinforce outdated or incorrect parametric knowledge, whereas accuracy gains require targeted, high-quality, task-specific data or retrieval.

Answer

Increasing the model's temperature to 1.5.

Answer

Using longer system prompts with multiple instructions.

Answer

Fine-tuning on a large corpus of general text.

Question 11

A company is deploying a generative AI system for medical diagnosis support. To comply with Google's AI Principles and regulatory requirements, which TWO actions are essential? (Select 2)

Accepted Answer

Implement a human-in-the-loop review for all diagnostic suggestions. Option A is correct because Google's AI Principles require that high-stakes applications such as medical diagnosis keep humans in the loop, ensuring a qualified clinician reviews and approves every AI-generated diagnostic suggestion before it affects patient care. Option C is correct because processing patient data for diagnosis support triggers legal obligations under GDPR and comparable privacy regulations, including lawful basis, data minimization, and protection of special-category health data. Option B is not essential here: a Model Card is useful documentation for transparency, but it is not a mandatory action for regulatory compliance in this scenario. Option D is not essential: SynthID watermarks AI-generated content to aid provenance, which is irrelevant to diagnostic decision support. Option E is not essential: using a larger model may improve accuracy but does not by itself satisfy AI Principles or regulatory requirements.

Answer

Publish a Model Card for the model

Answer

Use SynthID to watermark all output

Answer

Use a larger model to improve accuracy

Question 12

A company is deploying a generative AI chatbot for customer support. They want to ensure that the chatbot does not generate harmful content and that they can customize the safety thresholds. Which TWO features in Vertex AI should they use? (Select 2)

Accepted Answer

Custom safety settings (adjustable thresholds). Option B (Custom safety settings with adjustable thresholds) is correct because Vertex AI's generative AI safety controls let you configure per-category harm thresholds (for example, blocking more or less aggressively for hate speech, harassment, sexually explicit, and dangerous content), which directly satisfies the requirement to customize safety thresholds. Option D (Safety filters) is correct because these filters are the mechanism that actually evaluates model prompts and responses against harm categories and blocks or allows content, preventing the chatbot from generating harmful output. Together, safety filters provide the enforcement while custom safety settings provide the tunable thresholds the company wants. Option A (AutoML Tables) is incorrect because it is a tabular ML training/prediction service, not a content-safety feature. Option C (Model Cards) is incorrect because it documents model details, intended use, and evaluations but does not filter or block harmful content. Option E (People + AI Guidebook) is incorrect because it is a design guidance resource, not a runtime safety control in Vertex AI.

Answer

AutoML Tables

Answer

Model Cards

Answer

People + AI Guidebook

Question 13

A data scientist is using a large language model to generate product descriptions. The descriptions are often too verbose. Which parameter adjustment is most appropriate?

Accepted Answer

Increase the frequency penalty.. Increasing the frequency penalty reduces the likelihood of the model repeating the same phrases or ideas, which directly addresses verbosity by discouraging repetitive or overly detailed descriptions. This parameter penalizes tokens that have already appeared in the generated text, promoting more concise and varied output. Other adjustments like temperature or top-k affect randomness and diversity but do not specifically target repetition or length.

Answer

Decrease the top-k value.

Answer

Increase the max output tokens.

Answer

Decrease the temperature.

Question 14

A model generates responses that frequently repeat phrases or words. Which parameter adjustment is most likely to fix this?

Accepted Answer

Increase repetition penalty. Increasing the repetition penalty directly discourages the model from selecting tokens that have already appeared in the generated sequence, thereby reducing repetitive phrases or words. This parameter works by subtracting a fixed penalty from the logits of previously generated tokens before applying the softmax function, making them less likely to be chosen again.

Answer

Increase top_k

Answer

Increase temperature

Answer

Increase max output tokens

Question 15

A company is developing an AI-powered interview assistant that screens job applicants. The responsible AI team wants to ensure the model does not discriminate based on gender, race, or age. Which TWO practices should they implement?

Accepted Answer

Regularly evaluate the model's outputs for bias using intersectional test sets.. Option B is correct because regularly evaluating the model's outputs for bias using intersectional test sets allows the team to detect discrimination across combinations of protected attributes such as gender, race, and age, which is essential for responsible AI screening tools. Option E is correct because a diverse and representative training dataset that includes candidates from various demographics helps reduce sampling bias and enables the model to learn patterns that generalize fairly across groups. Option A is not appropriate because deploying without human oversight removes the ability to catch discriminatory or erroneous decisions, and consistency alone does not guarantee fairness. Option C is unrelated because SynthID watermarking is used to mark AI-generated content, not to mitigate bias in hiring models. Option D is not sufficient and can be harmful because simply removing demographic attributes does not ensure fairness and may allow proxy variables to perpetuate discrimination while also preventing the measurement of disparate impact.

Answer

Deploy the model without human oversight to ensure consistency.

Answer

Use SynthID watermarking on all model outputs.

Answer

Remove all demographic attributes from the training data to ensure fairness.