Courseiva

CCNA Fundamentals of Generative AI Questions

75 of 156 questions · Page 2/3 · Fundamentals of Generative AI · Answers revealed

76
MCQeasy

A developer wants to generate product descriptions from a list of features using Vertex AI. Which model type is best suited for this task?

A.An embedding model (e.g., textembedding-gecko@001).
B.A chat model (e.g., chat-bison@001).
C.A text generation model (e.g., text-bison@001).
D.A code generation model (e.g., code-bison@001).
AnswerC

A text generation model such as text-bison@001 maps a feature list to fluent prose, satisfying the stem's requirement to produce product descriptions. Unlike classification or embedding models, it outputs free-form natural language, which is precisely the generative capability Vertex AI needs here.

Why this answer

Text-bison@001 is a dedicated text generation model optimized for tasks like summarization, translation, and content creation from structured inputs. It can take a list of features as a prompt and generate coherent, descriptive product descriptions without needing conversational context or code-specific outputs.

Exam trap

The trap here is that candidates may confuse 'text generation' with 'chat' or 'embedding' models, assuming any generative model can handle the task, but Vertex AI separates these by specialization, and the exam tests awareness of which model class is purpose-built for non-conversational, non-code text creation.

How to eliminate wrong answers

Option A is wrong because embedding models like textembedding-gecko@001 are designed to convert text into numerical vectors for similarity search or clustering, not for generating new text. Option B is wrong because chat models like chat-bison@001 are optimized for multi-turn conversational interactions, not for single-turn structured generation tasks like producing descriptions from a feature list. Option D is wrong because code generation models like code-bison@001 are specialized for generating programming code, not natural language product descriptions.

77
Multi-Selectmedium

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.)

Select 2 answers
A.They are always smaller and cheaper to run than task-specific models because they do less work.
B.They must be trained from scratch for every new use case before they can produce any useful output.
C.They can generate novel content such as text, images, or code rather than only classifying existing inputs.
D.They typically require a large labeled dataset specific to each target task before they can generate text.
E.They are pretrained on broad data and can be adapted to many downstream tasks through prompting or fine-tuning.
AnswersC, E

Generative foundation models produce new content, including prose, summaries, images, and source code, instead of merely assigning a label to an input. This generative capability underpins use cases like drafting, translation, and code assistance. It distinguishes them from discriminative models whose output is a category or score.

Why this answer

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.

Exam trap

The trap here is mixing classical supervised-learning assumptions, such as needing large per-task labeled datasets, into descriptions of foundation models.

78
MCQhard

A financial services firm is designing a generative AI assistant for advisors. Compliance requires that every response be traceable to an approved source document and that no response rely on the model's pretrained knowledge alone. The team plans to use Gemini models on Vertex AI. Which design choice most directly satisfies the traceability requirement?

A.Enable grounding with citations against a curated corpus of approved documents
B.Fine-tune the model on historical advisor emails to match the firm's tone
C.Increase the model's temperature so responses vary and appear less templated
D.Set a low maximum output token limit to keep answers short and reviewable
AnswerA

Grounding with citations ties each generated statement to retrieved source passages and returns references the firm can audit. This directly satisfies the requirement that responses trace back to approved documents and avoid unsupported pretrained knowledge. Curating the corpus ensures only sanctioned material is retrievable, which is essential in a regulated advisory context.

Why this answer

Grounding with citations retrieves passages from a curated approved corpus and returns references alongside generated text, giving auditors a clear link from each claim to its source. This prevents reliance on pretrained knowledge alone. Temperature, fine-tuning, and token limits influence style, behavior, or length but cannot establish provenance, so they do not satisfy the compliance mandate.

Exam trap

The trap here is equating fine-tuning or brevity controls with traceability, when only retrieval-based grounding with citations provides auditable source references.

79
MCQeasy

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?

A.AutoML Natural Language
B.Vertex AI Gemini API
C.Dialogflow CX
D.Cloud Natural Language API
AnswerB

The Vertex AI Gemini API provides access to Google's Gemini models through a fully managed endpoint, abstracting infrastructure management. It supports text generation tasks like drafting personalized emails, and integrates with other Google Cloud services for security and scalability. This aligns with the requirement for minimal operational overhead.

Why this answer

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.

Exam trap

The trap here is confusing pre-trained NLP APIs that analyze text with generative APIs that create new text.

80
Multi-Selecthard

Which THREE of the following are key considerations when deploying a generative AI model in a production environment with strict latency requirements? (Choose three.)

Select 3 answers
A.Deploy the largest model variant available to ensure highest quality.
B.Implement speculative decoding to generate candidate tokens with a smaller draft model and verify with the large model.
C.Use model quantization (e.g., int8) to reduce precision and speed up matrix multiplications.
D.Cache the key-value caches from previous decoding steps to avoid redundant computation.
E.Increase the inference batch size to maximize GPU utilization.
AnswersB, C, D

Speculative decoding attacks latency directly: a smaller draft model proposes candidate tokens cheaply, and the large model verifies them in parallel, preserving output quality while cutting sequential decoding steps. This meets the strict latency requirement without retraining or accuracy loss.

Why this answer

Option B is correct because speculative decoding uses a smaller, faster draft model to propose multiple candidate tokens that the larger target model then verifies in parallel, reducing the number of expensive forward passes and lowering end-to-end latency while preserving output quality. Option C is correct because quantizing weights and activations to lower precision such as int8 shrinks memory bandwidth demands and enables faster matrix multiplications on supported hardware, directly cutting per-token inference time. Option D is correct because caching the key-value tensors from prior decoding steps avoids recomputing attention over the entire sequence for each new token, which is essential for keeping autoregressive generation latency low.

Option A is not appropriate because the largest model variant increases compute and memory cost, worsening latency rather than meeting strict requirements. Option E is not appropriate because increasing batch size improves throughput and GPU utilization but typically raises per-request latency, which conflicts with strict latency goals.

Exam trap

Google Cloud often tests the distinction between latency and throughput, so the trap here is that candidates confuse batch size (which improves throughput) with latency reduction, or assume larger models always yield better performance without considering inference speed.

81
MCQmedium

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?

A.Optimize the model using quantization and reduce the number of replicas
B.Switch to batch prediction instead of online prediction
C.Change the machine type to n1-standard-8 and enable autoscaling with min replicas=1, max replicas=5
D.Add a GPU accelerator to the existing machine
AnswerC

CPU is saturated above 90%, so the bottleneck is compute per replica. Doubling to n1-standard-8 gives each replica more vCPU headroom, and autoscaling to five replicas absorbs peak-hour concurrency while min replicas=1 keeps idle cost low.

Why this answer

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.

Exam trap

The trap here is that candidates often assume adding a GPU (Option D) is always the best way to reduce inference latency, but for CPU-bound models with high utilization, scaling out with more replicas and a larger CPU machine is more cost-effective and directly addresses the bottleneck.

How to eliminate wrong answers

Option A is wrong because quantization reduces model size and can improve latency, but reducing the number of replicas would worsen the bottleneck by decreasing capacity, not help with high CPU utilization. Option B is wrong because batch prediction is designed for asynchronous, non-real-time processing and does not solve online inference latency during peak hours; it would change the use case entirely. Option D is wrong because adding a GPU accelerator to the existing n1-standard-4 machine would not address the CPU bottleneck (inference is CPU-bound in this scenario) and would increase cost unnecessarily without guaranteed latency improvement for a CPU-bound model.

82
Multi-Selecthard

A team is designing prompts for a generative AI application that summarizes long legal contracts. They want outputs that are accurate, consistently formatted, and safe from leaking instructions. Which two prompt engineering practices should they apply? (Choose two.)

Select 2 answers
A.Set temperature to the maximum value to encourage more varied legal wording.
B.Instruct the model to ignore any instructions embedded inside the contract text.
C.Insert the entire contract text without delimiters and ask the model to figure out the important parts.
D.Provide a clear task instruction with an example of the desired output format.
E.Ask the model to reveal its system instructions so the team can verify them.
AnswersB, D

Contracts may contain clauses that look like instructions, and a malicious or accidental phrase could steer the model. Explicitly telling the model to treat document content as data and not as commands reduces prompt-injection risk and helps keep the summary faithful to the intended task, which is a recommended safety practice.

Why this answer

Giving a clear task plus a formatting example improves output consistency, and explicitly telling the model to ignore instructions inside the contract text reduces prompt-injection risk. These two practices together address accuracy, formatting stability, and safety when summarizing long legal documents, which were the team's stated goals.

Exam trap

The trap here is treating any instruction about instructions as helpful, when asking the model to reveal or obey embedded text can weaken safety instead of strengthening it.

83
MCQhard

A company is deploying a generative AI model on Vertex AI for real-time chat. They observe that the model sometimes generates toxic or biased responses. They want to implement a safety mechanism to filter out such content before it reaches users. Which Google Cloud feature should they use?

A.Sensitive Data Protection
B.Vertex AI Model Monitoring
C.Cloud Armor
D.Vertex AI Safety Filters
AnswerD

Vertex AI Safety Filters are designed to detect and block harmful content, including toxicity, bias, and explicit material. They can be configured with thresholds and applied to both prompts and responses in real time. This directly addresses the need to filter out toxic or biased outputs before they reach users.

Why this answer

Vertex AI Safety Filters are specifically built to detect and block harmful content in generative AI applications. They can be applied to both input and output, providing real-time protection. Other services like Cloud Armor or Sensitive Data Protection serve different purposes and cannot effectively filter toxic or biased language.

Exam trap

The trap here is confusing security or data privacy services with content safety filters designed for generative AI.

84
MCQeasy

A retail company wants to build an internal assistant that answers employee questions using the company's own HR policy documents. The team has no machine learning engineers and wants a managed Google Cloud approach that grounds responses in those documents without training a new foundation model. Which Google Cloud capability best fits this need?

A.Deploying an open-source model on a Compute Engine GPU VM and fine-tuning it weekly
B.Vertex AI Search with grounding on a data store built from the HR documents
C.Using BigQuery ML to run a logistic regression over HR ticket categories
D.Training a custom foundation model from scratch on the HR documents
AnswerB

Vertex AI Search lets teams index enterprise documents into a data store and then ground generative responses on retrieved passages. It is a managed service requiring no model training, and it directly addresses the requirement to answer from the company's own HR policies. This matches the no-ML-engineer constraint and the grounding goal precisely.

Why this answer

Vertex AI Search provides managed retrieval over enterprise documents and grounds generative answers in that content, so employees receive responses based on current HR policy rather than the model's general knowledge. It requires no foundation-model training and minimal ML operations, matching the team's skills and timeline. The other choices either demand heavy ML work or address unrelated prediction tasks.

Exam trap

The trap here is assuming grounding requires fine-tuning a model, when retrieval-augmented grounding through a managed search service is the intended no-training approach.

85
MCQeasy

A startup is building a customer service chatbot that generates responses in real-time. They want the model to have up-to-date information on the latest product catalog but cannot afford frequent fine-tuning. Which technique should they use to inject current data into the model without retraining?

A.Rely on the model's zero-shot capabilities to infer product details.
B.Use retrieval-augmented generation (RAG) to fetch relevant documents from a vector database at inference time.
C.Craft detailed system prompts that include the entire product catalog in the prompt.
D.Fine-tune the base model weekly on the latest product catalog.
AnswerB

RAG decouples knowledge from model weights: relevant catalogue documents are retrieved from a vector database and injected into the prompt at inference time. This satisfies the constraint of current data without the cost or delay of frequent fine-tuning.

Why this answer

Retrieval-Augmented Generation (RAG) is the correct technique because it allows the chatbot to fetch the most current product catalog entries from an external vector database at inference time, without requiring any model retraining. This keeps responses grounded in up-to-date information while avoiding the cost and latency of frequent fine-tuning.

Exam trap

Google Cloud often tests the distinction between in-context learning (via RAG or prompt engineering) and parametric knowledge (via fine-tuning), trapping candidates who think that simply adding more data to the prompt is scalable or that zero-shot inference can substitute for external retrieval.

How to eliminate wrong answers

Option A is wrong because zero-shot capabilities rely solely on the model's pre-existing knowledge, which cannot incorporate new or updated product catalog details without retraining. Option C is wrong because crafting detailed system prompts with the entire product catalog would exceed the model's context window limits and incur high token costs, making it impractical for real-time inference. Option D is wrong because fine-tuning weekly is expensive, time-consuming, and contradicts the requirement to avoid frequent retraining; it also risks catastrophic forgetting of previously learned information.

86
MCQmedium

A developer wants to build a RAG application using Vertex AI. Which vector database is natively integrated with Vertex AI for storing embeddings?

A.Firestore
B.Vertex AI Vector Search
C.Cloud SQL
D.Bigtable
AnswerB

Vertex AI Vector Search is Google Cloud's native vector store, purpose-built for the Vertex AI ecosystem, so embeddings generated by Vertex AI models can be indexed and queried without custom integration work. It satisfies the stem's requirement for a natively integrated vector database, unlike third-party options such as Pinecone or open-source alternatives.

Why this answer

Vertex AI Vector Search is the native vector database integrated with Vertex AI for storing and querying embeddings. It is purpose-built for high-dimensional vector similarity search, enabling efficient retrieval in RAG applications without requiring external infrastructure.

Exam trap

Google Cloud often tests the misconception that any database can store embeddings equally well, but the key differentiator is native vector indexing and ANN search support, which only Vertex AI Vector Search provides among the listed options.

How to eliminate wrong answers

Option A is wrong because Firestore is a NoSQL document database designed for storing structured data, not optimized for vector similarity search or embedding storage. Option C is wrong because Cloud SQL is a relational database service (MySQL, PostgreSQL, SQL Server) that lacks native vector indexing and similarity search capabilities required for RAG. Option D is wrong because Bigtable is a wide-column NoSQL database for large-scale analytical workloads, not designed for low-latency vector similarity queries.

87
MCQmedium

A company wants to build a chatbot that answers questions based on internal documents. Which approach is most appropriate?

A.Use a pre-trained model without any customizations
B.Train a custom model from scratch
C.Fine-tune a model on the documents
D.Use a prompt with the documents in the context
AnswerD

Placing the internal documents directly in the prompt context lets the model condition its answer on that retrieved text, grounding responses in company content without retraining. This suits document question answering where the corpus changes and fine-tuning would be impractical.

Why this answer

Retrieval-Augmented Generation (RAG) allows the chatbot to dynamically include relevant internal documents in the prompt context without modifying the underlying model. This approach leverages the pre-trained model's language understanding while grounding answers in specific, up-to-date internal data, avoiding the cost and latency of fine-tuning or retraining.

Exam trap

Google Cloud often tests the misconception that fine-tuning is the only way to incorporate proprietary data, but RAG is the most appropriate for dynamic, retrieval-based Q&A because it avoids retraining and keeps the model's knowledge current.

How to eliminate wrong answers

Option A is wrong because a pre-trained model without customization lacks access to the company's internal documents, leading to hallucinated or generic answers not grounded in proprietary data. Option B is wrong because training a custom model from scratch is computationally prohibitive and unnecessary; it requires massive labeled datasets and resources, whereas RAG achieves the same goal with far less effort. Option C is wrong because fine-tuning on documents teaches the model to memorize specific content, which is inefficient for large, frequently updated document sets and risks catastrophic forgetting, whereas RAG keeps the model static and retrieves fresh context per query.

88
Multi-Selectmedium

A healthcare startup is building a generative AI application to draft patient education materials. They need to ensure the outputs are accurate, up-to-date, and tailored to each patient's condition. Which two techniques should they use to ground the model's responses in reliable medical knowledge? (Choose two.)

Select 2 answers
A.Retrieval-augmented generation (RAG) using a curated medical knowledge base.
B.Fine-tuning the model on a large dataset of historical patient education materials.
C.Using a model with a larger context window to include more patient history.
D.Prompt engineering with few-shot examples of desired outputs.
E.Implementing a fact-checking layer that queries a trusted medical API for validation.
AnswersA, E

RAG retrieves relevant documents from a trusted knowledge base and provides them as context to the model, ensuring responses are grounded in authoritative, current information. This reduces hallucinations and allows tailoring to specific conditions by retrieving patient-specific or condition-specific guidelines. It is a core technique for building accurate generative AI applications in specialized domains.

Why this answer

Retrieval-augmented generation (RAG) grounds the model by providing relevant, trusted documents as context, while a fact-checking layer validates outputs against a reliable API. Together, they ensure accuracy and currency. Fine-tuning, few-shot prompting, and larger context windows do not inherently guarantee factual correctness or access to the latest medical knowledge.

Exam trap

The trap here is assuming that fine-tuning or a larger context window alone can ground a model in reliable knowledge, when retrieval and validation are needed for accuracy and currency.

89
MCQmedium

A retail company wants to use a generative AI model to create personalized product descriptions for thousands of items. They need the descriptions to be consistent in style and format, but they also want to avoid the model inventing false features. Which approach best balances consistency and accuracy?

A.Use a high temperature to encourage creativity and let the model generate unique descriptions for each item.
B.Increase the max output tokens to allow the model to elaborate more on each product.
C.Fine-tune the model on a dataset of existing product descriptions to learn the style.
D.Set a low temperature and provide a structured prompt with clear guidelines and examples.
AnswerD

A low temperature reduces randomness, making outputs more deterministic and consistent. A structured prompt with guidelines and examples further constrains the model to follow the desired style and format. This combination minimizes the risk of hallucinated features while maintaining uniformity across thousands of descriptions, directly addressing both requirements.

Why this answer

Setting a low temperature makes the model's output more predictable and consistent, while a structured prompt with guidelines and examples steers the model to follow a specific style and format. This combination reduces the likelihood of hallucinated features because the model is constrained by explicit instructions and examples. Other approaches either increase variability or do not directly address accuracy.

Exam trap

The trap here is thinking that fine-tuning alone ensures accuracy, when in fact prompt design and temperature settings are more direct controls for consistency and hallucination reduction.

90
MCQhard

A team is building a medical diagnosis assistant using a foundation model. To comply with regulations, they need to ensure the model does not make up facts. What is the best approach?

A.Use a small model to hallucinate less
B.Use grounding with Vertex AI Search
C.Reduce temperature to 0
D.Fine-tune on medical journals
AnswerB

Grounding with Vertex AI Search anchors responses to retrieved, authoritative sources, so the model cites verified content instead of fabricating facts. This directly satisfies the regulatory requirement that the medical assistant must not make up information.

Why this answer

Grounding with Vertex AI Search is the best approach because it connects the foundation model's outputs to a verifiable, curated knowledge base, ensuring factual accuracy and compliance with regulations that prohibit hallucination. By retrieving information from a trusted source (e.g., medical databases) in real time, the model can cite evidence and avoid generating unverified claims.

Exam trap

Google Cloud often tests the misconception that reducing temperature or using a smaller model can eliminate hallucination, when in fact only grounding with external, verifiable data sources can reliably prevent fact fabrication in high-stakes domains.

How to eliminate wrong answers

Option A is wrong because using a smaller model does not inherently reduce hallucination; smaller models have less capacity and may actually hallucinate more due to limited training data and weaker reasoning. Option C is wrong because reducing temperature to 0 makes the model deterministic but does not prevent it from generating plausible-sounding but false information; it still relies on its parametric knowledge, which can be incomplete or outdated. Option D is wrong because fine-tuning on medical journals alone does not guarantee factual accuracy; the model may memorize and reproduce errors, and it cannot dynamically verify facts against a live, authoritative source.

91
MCQeasy

A marketing team wants to generate product descriptions using a text generation model on Vertex AI. They need consistent output style across all descriptions, including tone and length. They have a small set of 10 high-quality example descriptions that capture the desired style. The team has limited ML expertise and wants a quick solution that does not require model retraining. Which approach should they use?

A.Use a pre-built template with no model input.
B.Fine-tune the model on a large external dataset of product descriptions.
C.Use few-shot prompting with the examples in the prompt.
D.Set the temperature to 0.9 to maximize creativity.
AnswerC

Few-shot prompting embeds the ten exemplar descriptions directly in the prompt, steering tone and length without weight updates. This satisfies the no-retraining constraint and suits limited ML expertise, since no training pipeline or labelled dataset is needed. It reliably enforces the consistent style the marketing team requires.

Why this answer

Few-shot prompting is the correct approach because it allows the team to inject the desired style, tone, and length directly into the prompt using the 10 high-quality examples, without any model retraining. This technique leverages the in-context learning capability of large language models on Vertex AI, enabling consistent output from a small set of demonstrations. It is ideal for teams with limited ML expertise as it requires only prompt engineering, not fine-tuning or infrastructure changes.

Exam trap

Google Cloud often tests the misconception that higher temperature always improves output quality, but the trap here is that temperature controls randomness, not consistency, so candidates may incorrectly choose Option D without understanding that low temperature is required for reproducible style and length.

How to eliminate wrong answers

Option A is wrong because a pre-built template with no model input cannot generate dynamic, context-aware product descriptions; it produces static text that lacks the flexibility and nuance of a generative model. Option B is wrong because fine-tuning on a large external dataset would require significant ML expertise, data preparation, and compute resources, contradicting the requirement for a quick solution without model retraining. Option D is wrong because setting temperature to 0.9 maximizes randomness and creativity, which is the opposite of what is needed for consistent output style; a lower temperature (e.g., 0.2) would be more appropriate for deterministic, reproducible results.

92
Multi-Selectmedium

A marketing team is using a generative AI model to create ad copy. They want to ensure the outputs align with their brand voice and avoid offensive language. Which two practices should they adopt? (Choose two.)

Select 2 answers
A.Set the model's temperature to the maximum value to encourage creativity.
B.Deploy multiple model versions and manually select the best output for each ad.
C.Provide clear instructions and examples in the prompt to guide the model's tone and style.
D.Fine-tune the model on a large dataset of competitor ads.
E.Use Vertex AI Safety Filters to block harmful content.
AnswersC, E

Prompt engineering with explicit instructions and examples helps steer the model toward the desired brand voice and away from offensive language. By specifying tone, style, and constraints, the team can influence the output effectively. This is a direct and flexible method to align generated content with brand guidelines without retraining the model.

Why this answer

To align outputs with brand voice and avoid offensive language, the team should use prompt engineering to guide tone and style, and implement Vertex AI Safety Filters to block harmful content. These two practices work together to ensure both stylistic and safety requirements are met. Other options either introduce risks or are inefficient.

Exam trap

The trap here is assuming that fine-tuning on competitor data or maximizing creativity will achieve brand alignment, when they actually increase risk and inconsistency.

93
MCQhard

A financial institution deploys a chatbot using Gemini Pro in Vertex AI. Compliance requires logging all user inputs and model outputs for audit. Which approach meets this requirement?

A.Capture logs via Cloud Monitoring
B.Enable Vertex AI Endpoint request-response logging
C.Use Cloud Logging sink with a filter for Vertex AI requests
D.Enable Vertex AI Model Registry logging
AnswerB

Endpoint request-response logging captures the full prompt and completion payloads for every prediction, storing them in Cloud Logging for audit retrieval. This directly satisfies the compliance constraint to log all user inputs and model outputs without altering application code.

Why this answer

Vertex AI Endpoint request-response logging captures both the user's input prompt and the model's generated output, which is precisely what compliance auditing requires. This feature logs the exact payloads sent to and received from the deployed model, ensuring a complete audit trail without additional configuration.

Exam trap

The trap here is that candidates confuse Cloud Logging sinks or Cloud Monitoring with the specific Vertex AI feature that must be explicitly enabled on the endpoint, assuming that default logging captures request-response payloads when it does not.

How to eliminate wrong answers

Option A is wrong because Cloud Monitoring is designed for metrics, alerts, and dashboards, not for capturing detailed request-response payloads for audit compliance. Option C is wrong because a Cloud Logging sink with a filter can only export logs that already exist; it does not enable the capture of Vertex AI request-response logs, which must be explicitly enabled on the endpoint. Option D is wrong because Vertex AI Model Registry logging tracks model version metadata and lifecycle events, not the user inputs and model outputs from inference calls.

94
MCQhard

A multimodal generative AI system processes both image and text inputs to produce captions. During inference, the image encoder sometimes produces noisy or missing features. Which architectural design decision best handles such input degradation without retraining?

A.Train a separate variational autoencoder to produce a clean latent representation from the noisy image.
B.Increase the image encoder’s capacity to better extract robust features.
C.Apply standard image preprocessing (e.g., denoising) to all inputs before feeding to the encoder.
D.Introduce a gating mechanism that learns to weigh image features based on confidence scores from the encoder.
AnswerD

A learned gating mechanism dynamically weights image features by encoder confidence, letting the decoder down-weight noisy or missing inputs at inference. This handles degradation without retraining, satisfying the stem's constraint of robustness to unreliable visual features.

Why this answer

A gating mechanism dynamically adjusts the contribution of image features based on confidence scores from the encoder, allowing the model to gracefully handle noisy or missing features without retraining. This architectural design learns to suppress unreliable image inputs and rely more on text or other modalities, ensuring robust caption generation under input degradation.

Exam trap

Google Cloud often tests the misconception that preprocessing or model capacity adjustments are the only ways to handle input noise, but the key insight is that architectural mechanisms like gating can adaptively handle degradation at inference time without retraining.

How to eliminate wrong answers

Option A is wrong because training a separate variational autoencoder (VAE) to produce clean latent representations requires additional training data and retraining, which contradicts the 'without retraining' constraint; it also adds complexity without addressing dynamic degradation during inference. Option B is wrong because increasing the image encoder’s capacity does not inherently handle noisy or missing features—it may overfit to training data and still produce unreliable outputs when inputs degrade, and it requires retraining to change capacity. Option C is wrong because standard image preprocessing like denoising is a fixed, non-adaptive approach that cannot compensate for missing features or varying noise levels, and it may discard useful information; it also does not leverage the model’s ability to learn confidence-based weighting.

95
MCQmedium

A marketing team uses a generative AI model to produce ad copy. They notice that when they set the temperature parameter to 0.1, the outputs are very similar across runs, but when they set it to 0.9, the outputs vary widely. They want a balance between creativity and consistency for a campaign that requires some variation but also brand alignment. Which temperature value should they choose?

A.0.5
B.0.0
C.2.0
D.1.0
AnswerA

A moderate temperature like 0.5 balances randomness and coherence. It allows the model to explore alternatives without drifting too far from likely outputs, which suits a campaign needing both variation and brand consistency. This setting is a common middle ground between deterministic and highly random generation.

Why this answer

A temperature of 0.5 provides a middle ground between deterministic and highly random outputs. It allows for some creativity and variation while keeping the text coherent and aligned with brand guidelines. Lower values like 0.0 reduce variation, while higher values like 1.0 or 2.0 increase randomness and risk off-brand content.

Exam trap

The trap here is assuming that higher temperature always yields better creativity without considering the trade-off in coherence and brand alignment.

96
MCQhard

Refer to the exhibit. A developer sees this error when trying to deploy a model from Vertex AI Model Registry. What is the most likely cause?

A.The region is not supported
B.The developer used the model display name instead of the full resource name
C.The model is not published
D.The model is in a different project
AnswerB

Deployment APIs require the fully qualified resource name, formatted as projects/{project}/locations/{location}/models/{model}. Supplying only the display name fails resource resolution, producing the error, since display names are not unique identifiers within the Vertex AI Model Registry.

Why this answer

The error occurs because Vertex AI Model Registry requires the full resource name (e.g., 'projects/{project}/locations/{region}/models/{model_id}') to deploy a model, not just the display name. The display name is a human-readable label that is not unique within a project, while the full resource name uniquely identifies the model version. Using the display name causes the API to fail with a 'not found' or 'invalid argument' error.

Exam trap

Google Cloud often tests the distinction between display names (non-unique, human-readable) and resource names (unique, API-required) in cloud services like Vertex AI, where candidates mistakenly assume display names can be used interchangeably with resource identifiers.

How to eliminate wrong answers

Option A is wrong because Vertex AI supports model deployment in all regions where the service is available, and the error message does not indicate a regional restriction. Option C is wrong because a model can be deployed from the registry even if it is not published to the public; publishing is only required for sharing with external users or making it available in the Model Garden. Option D is wrong because the error would reference a cross-project permission issue (e.g., 'permission denied' or 'resource not found in project'), not a display name mismatch.

97
Multi-Selecthard

A product team is designing a generative AI application to create personalized email campaigns. They want to ensure the model produces high-quality, relevant content while minimizing risks such as bias and harmful outputs. Which two practices should the team implement? (Choose two.)

Select 2 answers
A.Fine-tune the model on a small dataset of previously successful email campaigns without any validation.
B.Implement content filters and safety thresholds to block harmful or inappropriate outputs.
C.Conduct regular fairness evaluations of the model's outputs across different demographic groups.
D.Disable all logging and monitoring to protect user privacy and reduce overhead.
E.Set the model's temperature to the maximum value to encourage diverse and creative emails.
AnswersB, C

Content filters and safety thresholds automatically detect and suppress toxic, biased, or otherwise harmful text before it reaches recipients. This is a critical layer of defense in generative AI applications, especially for customer-facing campaigns. It helps maintain brand safety and compliance with regulations, directly addressing the goal of minimizing harmful outputs.

Why this answer

Regular fairness evaluations help detect and reduce bias across demographic groups, ensuring equitable and appropriate content. Content filters and safety thresholds act as a safeguard to block harmful or inappropriate outputs before they reach users. Together, these practices address both bias and harmful content, aligning with responsible AI development.

Other options either increase randomness, risk overfitting, or remove necessary oversight.

Exam trap

The trap here is focusing only on creative output and overlooking the need for systematic bias detection and automated safety filters.

98
MCQeasy

A retail company wants to use a generative AI model to create unique product descriptions for thousands of items. They need the model to produce varied, human-like text without being explicitly programmed for each product. Which core capability of generative AI does this scenario primarily rely on?

A.Forecasting future sales trends
B.Clustering similar products based on features
C.Generating novel content based on patterns learned from training data
D.Classifying products into predefined categories
AnswerC

Generative AI models learn statistical patterns from large datasets and can produce new, original content such as text, images, or code. In this scenario, the model generates unique product descriptions by leveraging its training on diverse text, without explicit programming for each item. This is the fundamental capability that enables creative and varied output.

Why this answer

Generative AI is designed to create new content by learning patterns from existing data. In this scenario, the model generates unique product descriptions without explicit programming, which is the essence of generative AI. The other options describe discriminative or predictive tasks that do not produce novel text, so they fail to meet the company's need for varied, human-like descriptions.

Exam trap

The trap here is confusing generative AI with traditional discriminative models that only classify or predict, rather than create new content.

99
MCQeasy

A company is using Vertex AI to deploy a text generation model for a chatbot. They want to reduce the response latency. Which configuration change is most effective?

A.Enable model quantization
B.Use a smaller model variant
C.Increase the number of GPUs
D.Use a larger batch size
AnswerB

A smaller model variant reduces the number of parameters processed per token, cutting inference compute and therefore response latency. This directly satisfies the stem's constraint of lowering latency for the Vertex AI chatbot, since generation time scales with model size rather than with prompt formatting or endpoint region.

Why this answer

Using a smaller model variant directly reduces the number of parameters and computational operations required per inference, which lowers latency. In Vertex AI, smaller models like `text-bison@002` have fewer layers and attention heads than larger counterparts, resulting in faster token generation without requiring hardware changes.

Exam trap

Google Cloud often tests the misconception that increasing compute resources (GPUs) or batch size always reduces latency, when in fact these optimizations target throughput, not per-request response time.

How to eliminate wrong answers

Option A is wrong because model quantization (e.g., reducing weights from FP32 to INT8) can reduce memory footprint and improve throughput, but it does not guarantee lower latency per request and may introduce accuracy trade-offs; it is not the most effective single change for latency reduction. Option C is wrong because increasing the number of GPUs can improve throughput for batch processing but does not reduce per-request latency; in fact, it may increase communication overhead and cost without speeding up individual inference. Option D is wrong because using a larger batch size increases throughput for concurrent requests but actually increases the latency for each individual request, as the model processes more sequences together before returning results.

100
Multi-Selecteasy

Which TWO are components of the Vertex AI Generative AI Studio?

Select 2 answers
A.Dataflow
B.Model Garden
C.Pipeline templates
D.Cloud Functions
E.Prompt Editor
AnswersB, E

Model Garden is a catalogue within Vertex AI Generative AI Studio for discovering, testing and deploying foundation models, including Google's own and third-party options. It is one of the studio's core components, satisfying the question's requirement.

Why this answer

Model Garden (B) is a core component of Vertex AI Generative AI Studio, providing a curated catalog of foundation models from Google and third parties (e.g., PaLM, Gemini, Llama) that users can browse, test, and deploy. Prompt Editor (E) is also a built-in Studio component that lets users interactively design, test, and refine prompts against foundation models without writing code. Dataflow (A) is a managed Apache Beam service for batch and streaming data pipelines, unrelated to the Studio's model/prompt tooling.

Pipeline templates (C) belong to Vertex AI Pipelines for orchestrating ML workflows, not the Generative AI Studio interface. Cloud Functions (D) is a serverless compute service for event-driven code, not a Generative AI Studio component.

Exam trap

Google Cloud often tests the distinction between core generative AI studio components (like Model Garden and Prompt Editor) and broader GCP services (like Dataflow or Cloud Functions) that are not part of the studio, leading candidates to select familiar but incorrect options.

101
Multi-Selectmedium

Which TWO are benefits of using retrieval-augmented generation (RAG) over fine-tuning?

Select 2 answers
A.No need for training
B.Higher accuracy on all tasks
C.More up-to-date information
D.Reduced model size
E.Lower latency
AnswersA, C

RAG eliminates the training step entirely: it supplies relevant context at inference time via a retrieval index, so the model weights stay frozen. This satisfies the stem's benefit of avoiding the compute, cost and labelled-data overhead that fine-tuning demands, letting you update knowledge by refreshing documents instead of retraining.

Why this answer

Option A is correct because RAG does not modify the model's weights; it retrieves relevant documents at inference time and injects them into the prompt, so no training or fine-tuning run is required to add new knowledge. Option C is correct because RAG can pull from an external, continuously updated knowledge base or index, letting the system answer with information newer than the model's static training cutoff, whereas fine-tuning bakes in knowledge only as of the training data used. Options B, D, and E are not benefits guaranteed by RAG: it does not yield higher accuracy on all tasks (retrieval quality can introduce errors), it does not reduce the underlying model's size since the same LLM is still used, and it typically adds retrieval overhead that can increase, not lower, latency.

Exam trap

Google Cloud often tests the misconception that RAG reduces latency or model size, when in fact it increases system complexity and inference time due to the retrieval step, while fine-tuning keeps the model unchanged in size and latency.

102
MCQeasy

A company wants to build a chatbot that answers questions using their internal knowledge base. Which approach is most suitable?

A.Use Retrieval-Augmented Generation (RAG)
B.Fine-tune a model on the knowledge base
C.Train a new model from scratch
D.Use zero-shot prompting with no context
AnswerA

RAG retrieves relevant context and generates answers, perfect for knowledge base Q&A.

Why this answer

RAG retrieves relevant passages from the internal knowledge base at query time and injects them into the model's context, so the chatbot answers with grounded, up-to-date, source-specific information without retraining. It is the standard pattern when the knowledge base changes frequently and citations or freshness matter. Fine-tuning, training from scratch, or zero-shot prompting do not reliably surface private, changing content.

Exam trap

The trap is assuming fine-tuning is the default way to add private knowledge; exams test whether you know RAG is preferred for dynamic, factual, source-grounded retrieval.

How to eliminate wrong answers

Option B is wrong because fine-tuning bakes knowledge into weights, which is expensive, slow to refresh, and prone to hallucination when facts change; it is better for style/format adaptation than factual retrieval. Option C is wrong because training a new model from scratch requires massive compute, data, and expertise, and still would not guarantee retrieval of current internal documents. Option D is wrong because zero-shot prompting with no context gives the model no access to the internal knowledge base, so answers will be generic or fabricated.

103
MCQmedium

Refer to the exhibit. A data scientist runs the gcloud command and sees the model listed. However, when they try to deploy the model to an endpoint, they get an error: 'Model is not deployable'. What is the most likely reason?

A.The model is still in training and not yet ready.
B.The model was imported from a custom container but without a serving specification or artifact.
C.The model does not have the correct IAM permissions assigned to the deployment service account.
D.The region for the endpoint is different from the model's region.
AnswerB

Importing a custom container without a serving specification or artifact leaves Vertex AI no defined prediction routine or weights to load, so the model registers but cannot be deployed. The serving specification is what makes a model deployable to an endpoint.

Why this answer

A model imported from a custom container must include a serving specification (e.g., a `predict` route) and an artifact (e.g., a saved model file) to be deployable. Without these, Vertex AI cannot determine how to serve predictions, resulting in the 'Model is not deployable' error. The `gcloud` command listing the model only confirms its registration, not its readiness for deployment.

Exam trap

Google Cloud often tests the misconception that a model listed in the registry is automatically deployable, but the trap here is that Vertex AI separates model registration from deployment readiness, requiring explicit serving configuration for custom containers.

How to eliminate wrong answers

Option A is wrong because if the model were still in training, it would not appear in the model list via `gcloud`; Vertex AI only registers a model after training completes. Option C is wrong because IAM permissions affect the deployment action itself (e.g., who can deploy), not the deployability status of the model; the error 'Model is not deployable' is a model-level validation, not an authorization failure. Option D is wrong because region mismatch between the endpoint and model would cause a resource-location error, not a 'Model is not deployable' error; Vertex AI enforces regional consistency but does not block deployment based on region alone.

104
Multi-Selectmedium

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

Select 2 answers
A.Provide context and examples in the prompt
B.Append random noise to prompts to improve creativity
C.Use clear and specific instructions
D.Always use the maximum possible number of tokens
E.Use negative prompts to discourage undesired outputs
AnswersA, C

Supplying context and few-shot examples steers the model toward the desired output format and domain, reducing ambiguity. It satisfies the best-practise criterion by grounding responses in relevant information rather than relying on the model's parametric memory alone.

Why this answer

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.

105
Multi-Selectmedium

A team is preparing to use Vertex AI Studio to prototype a generative AI application. They want to understand which factors directly influence the quality and safety of the model's responses. (Choose two.)

Select 2 answers
A.The configured safety filters and their threshold settings
B.The color scheme of the application's user interface
C.The physical region where the developer's laptop is located
D.The programming language used to call the Vertex AI API
E.The clarity and specificity of the prompt provided by the user
AnswersA, E

Safety filters with adjustable thresholds determine which categories of content the model will block or allow, directly affecting the safety of responses. Tightening thresholds reduces harmful output, while loosening them permits more content. These settings are a core control for responsible deployment and must be tuned to the application's risk profile.

Why this answer

Prompt construction and safety filter configuration are the two levers that most directly shape a generative model's responses in Vertex AI Studio. A precise prompt improves relevance and accuracy, while safety thresholds govern what content is permitted. Presentation choices, client location, and programming language affect the surrounding application but not the model's generation or filtering behavior, so they do not influence response quality or safety.

Exam trap

The trap here is assuming that implementation details like the calling language or UI design affect model output, when only prompt and safety configuration do.

106
MCQeasy

A data scientist needs to fine-tune a foundation model for a sentiment analysis task without managing infrastructure. Which Google Cloud service should they use?

A.Compute Engine
B.BigQuery ML
C.Cloud Run
D.Vertex AI Model Garden
AnswerD

Vertex AI Model Garden supplies foundation models with tuning workflows that run on managed infrastructure, so no cluster provisioning is needed. It directly satisfies the stem's constraint of fine-tuning for sentiment analysis without infrastructure management, unlike raw compute or self-hosted alternatives.

Why this answer

Vertex AI Model Garden is the correct service because it provides a curated hub of foundation models that can be fine-tuned with managed infrastructure, eliminating the need for the data scientist to provision or manage servers. It supports one-click deployment and fine-tuning workflows for sentiment analysis, directly addressing the requirement to avoid infrastructure management.

Exam trap

The trap here is that candidates often confuse BigQuery ML's ability to train models on tabular data with the capability to fine-tune large language models, but BigQuery ML does not support fine-tuning of foundation models for NLP tasks.

How to eliminate wrong answers

Option A is wrong because Compute Engine is an IaaS offering that requires the user to manually provision, configure, and manage virtual machines, which contradicts the requirement of not managing infrastructure. Option B is wrong because BigQuery ML is designed for creating and executing machine learning models using SQL queries on structured data in BigQuery, not for fine-tuning large foundation models for natural language tasks like sentiment analysis. Option C is wrong because Cloud Run is a serverless container platform for running stateless HTTP-driven applications, but it does not provide native support for fine-tuning foundation models; it would require the user to build and manage the fine-tuning pipeline themselves.

107
MCQmedium

A company fine-tunes a model using Vertex AI and notices the model's performance drops on the original training task (e.g., language understanding) after fine-tuning for a new task (e.g., summarization). What could be the cause?

A.Data leakage
B.Model quantization
C.Catastrophic forgetting
D.Underfitting
AnswerC

Catastrophic forgetting occurs when gradient updates for the new summarisation task overwrite the weights encoding the original language-understanding capability. This directly explains the performance drop on the earlier task, satisfying the stem's observation of degraded prior-task accuracy after fine-tuning.

Why this answer

Catastrophic forgetting occurs when a neural network loses previously learned knowledge upon being fine-tuned on a new task. In this scenario, fine-tuning the model for summarization overwrites the weights responsible for language understanding, causing performance degradation on the original task. This is a well-known limitation of sequential fine-tuning in deep learning.

Exam trap

Google Cloud often tests the distinction between catastrophic forgetting and underfitting, as candidates may mistakenly think the model simply didn't learn the new task well, rather than recognizing that it forgot the original task due to weight overwriting.

How to eliminate wrong answers

Option A is wrong because data leakage refers to the inadvertent exposure of target information during training, which would typically inflate performance metrics rather than cause a drop on the original task. Option B is wrong because model quantization reduces numerical precision (e.g., from FP32 to INT8) to improve inference speed and memory efficiency, but it does not inherently cause performance loss on a previously learned task; any accuracy loss from quantization is generally uniform across tasks. Option D is wrong because underfitting means the model fails to capture patterns in the training data, resulting in poor performance on both the original and new tasks, not a selective drop on the original task after fine-tuning.

108
MCQhard

A developer uses the Vertex AI Python SDK to call a Gemini model for structured JSON output. However, the model often returns malformed JSON. Which parameter should the developer set in the generation configuration to enforce valid JSON output?

A.Set the temperature to a lower value (0.1) to reduce variation.
B.Set the 'response_mime_type' parameter to 'application/json'.
C.Include few-shot examples of the desired JSON format in the system prompt.
D.Switch to a smaller model to reduce complexity.
AnswerB

Setting `response_mime_type` to `application/json` constrains Gemini's decoding to emit syntactically valid JSON, directly satisfying the stem's requirement for structured output. This parameter enforces the output format at generation time, eliminating the malformed responses the developer currently receives from the Vertex AI Python SDK.

Why this answer

Setting `response_mime_type` to `'application/json'` in the generation configuration instructs the Gemini API to constrain the model's output to valid JSON format. This parameter leverages the model's native structured output capability, ensuring the response adheres to JSON syntax without relying on post-processing or prompt engineering.

Exam trap

Google Cloud often tests the misconception that prompt engineering (e.g., few-shot examples or temperature tuning) can reliably enforce structured output, when in fact the correct approach is to use the API's native structured output parameter like `response_mime_type`.

How to eliminate wrong answers

Option A is wrong because lowering temperature reduces randomness but does not enforce structural constraints; the model can still produce malformed JSON due to token-level deviations. Option C is wrong because few-shot examples in the system prompt improve formatting consistency but do not guarantee valid JSON output, as the model may still generate syntax errors or deviate from the schema. Option D is wrong because switching to a smaller model reduces capacity and may increase the likelihood of malformed output, and model size does not address the need for structured output enforcement.

109
MCQmedium

A company wants to use a generative AI model to create product descriptions from a list of features. They need the model to consistently follow a specific format: a headline, followed by three bullet points, and a closing sentence. Which technique should they use to guide the model's output structure?

A.Increasing the temperature to allow more creative freedom.
B.Using a larger model with more parameters.
C.Setting top-k to 1 to force deterministic output.
D.Few-shot prompting with examples of the desired format.
AnswerD

Few-shot prompting provides the model with several examples of the input-output format, allowing it to learn the pattern and apply it to new inputs. This is effective for enforcing a consistent structure like headline, bullets, and closing. It leverages in-context learning without retraining the model.

Why this answer

Few-shot prompting is ideal for teaching the model a specific output format by providing examples. It allows the model to infer the pattern and apply it consistently. Other options like increasing temperature or changing model size do not directly address the need for structured output.

Exam trap

The trap here is assuming that a larger model or deterministic settings will automatically produce the desired format, when prompt design is the key.

110
Multi-Selecthard

Which THREE of the following are potential risks when deploying generative AI?

Select 4 answers
A.Hallucinations
B.Memorization of sensitive training data
C.Bias and fairness issues
D.Increased model accuracy
E.Toxic or harmful content generation
AnswersA, B, C, E

Hallucinations are a genuine generative AI risk: the model generates fluent but factually incorrect or fabricated output because it predicts probable tokens rather than verifying truth. This directly satisfies the stem's requirement to identify a deployment risk, since unverified content can mislead users and damage trust.

Why this answer

Option A (Hallucinations) is correct because generative AI models can produce fluent but factually incorrect or fabricated outputs, which poses a risk when users trust them for decisions or information. Option B (Memorization of sensitive training data) is correct because models may reproduce verbatim or near-verbatim content from their training corpus, potentially leaking PII, proprietary data, or copyrighted material. Option C (Bias and fairness issues) is correct because models learn statistical patterns from training data and can amplify societal biases, leading to discriminatory or unfair outcomes in hiring, lending, or content moderation.

Option E (Toxic or harmful content generation) is correct because generative models can be prompted or inadvertently produce hate speech, harassment, self-harm instructions, or other harmful material. Option D (Increased model accuracy) is not a risk but a potential benefit, so it does not belong among the deployment risks.

Exam trap

Google Cloud often tests the distinction between risks and benefits, so the trap here is that candidates may mistakenly identify 'increased model accuracy' as a risk, when it is actually a performance improvement and not a deployment risk.

111
MCQeasy

A startup is building a customer support chatbot using Vertex AI and wants to ground responses in their product documentation to reduce hallucinations. Which approach should they use?

A.Enable Vertex AI Grounding with a custom enterprise data store containing the documentation.
B.Use the Codey API for text generation.
C.Use the base model without any grounding to maximize flexibility.
D.Fine-tune the model on the documentation and deploy.
AnswerA

Grounding with a custom enterprise data store retrieves passages from the product documentation and injects them into the prompt, so responses cite actual content rather than relying on parametric memory. This directly addresses the hallucination constraint in the stem.

Why this answer

Vertex AI Grounding with a custom enterprise data store is the correct approach because it allows the chatbot to retrieve and cite specific chunks from the product documentation in real time, directly reducing hallucinations by constraining responses to verified content. This method uses the underlying grounding service to query a vector-based data store (powered by Vertex AI Search) and append source references to the model's output, ensuring factual accuracy without retraining.

Exam trap

Google Cloud often tests the misconception that fine-tuning is the best way to incorporate domain knowledge, but the trap here is that fine-tuning does not provide dynamic, verifiable grounding with citations, whereas Vertex AI Grounding with a custom data store does, making it the correct choice for reducing hallucinations in a retrieval-augmented generation use case.

How to eliminate wrong answers

Option B is wrong because the Codey API is designed for code generation tasks (e.g., code completion, chat), not for grounding responses in external documents; it lacks the retrieval-augmented generation (RAG) capabilities needed to reduce hallucinations from product documentation. Option C is wrong because using a base model without grounding maximizes flexibility but also maximizes the risk of hallucination, as the model relies solely on its training data and cannot verify facts against the documentation. Option D is wrong because fine-tuning the model on the documentation embeds the content into the model's weights, which is static, costly to update, and does not provide real-time citation or retrieval; it also risks overfitting and does not leverage Vertex AI's built-in grounding infrastructure for dynamic fact-checking.

112
MCQeasy

A prompt engineer wants to improve the model's adherence to a specific output format (e.g., always start with a greeting). Which technique should they try first?

A.Use a lower temperature to make the output more deterministic.
B.Fine-tune the model on many examples of the desired format.
C.Include a system instruction at the beginning of the prompt that specifies the desired format.
D.Modify the model's tokenizer to encode the format rules.
AnswerC

System instructions are processed as high-priority context that conditions every response, making them the most direct control for enforcing a consistent output format such as an opening greeting. This precedes few-shot examples or post-processing in effort and reliability.

Why this answer

System instructions are the most direct and efficient method to enforce output formatting in large language models. By placing a clear directive at the beginning of the prompt (e.g., 'Always start your response with a greeting'), the model's attention mechanism is guided to prioritize this rule during generation, without requiring retraining or hyperparameter changes.

Exam trap

Google Cloud often tests the misconception that hyperparameter tuning (like temperature) can enforce structural output rules, when in fact it only controls randomness, not format adherence.

How to eliminate wrong answers

Option A is wrong because lowering temperature reduces randomness but does not enforce a specific structural rule like starting with a greeting; it only makes token selection more deterministic, which could still produce varied formats. Option B is wrong because fine-tuning is a resource-intensive process that requires a curated dataset and retraining, making it an overkill for a simple formatting constraint that can be achieved with a prompt instruction. Option D is wrong because modifying the tokenizer would alter how input text is split into tokens, not how the model adheres to output format rules; tokenizers have no mechanism to enforce generation constraints.

113
MCQeasy

Refer to the exhibit. A developer sees this error when trying to call a Vertex AI endpoint for online prediction. What permission does the requesting identity need to be granted?

A.aiplatform.prediction.predict
B.aiplatform.endpoints.predict
C.aiplatform.endpoints.use
D.aiplatform.models.predict
AnswerB

Granting `aiplatform.endpoints.predict` satisfies the online prediction constraint: it authorises the requesting identity to send prediction requests to a Vertex AI endpoint. This permission is contained in roles such as Vertex AI User, and is the minimum required for calling `predict` on a deployed endpoint.

Why this answer

The error occurs when calling a Vertex AI endpoint for online prediction, which requires the `aiplatform.endpoints.predict` permission. This permission is specifically scoped to the endpoint resource, allowing the identity to send prediction requests to a deployed model endpoint. The correct IAM role binding must include this permission for the requesting identity to successfully invoke the endpoint.

Exam trap

Google Cloud often tests the distinction between permissions scoped to endpoints versus models, and candidates mistakenly choose `aiplatform.models.predict` because they think prediction is always tied to the model, not the endpoint serving it.

How to eliminate wrong answers

Option A is wrong because `aiplatform.prediction.predict` is not a valid IAM permission in Vertex AI; the correct permission for prediction is scoped to the endpoint or model resource, not a generic 'prediction' service. Option C is wrong because `aiplatform.endpoints.use` does not exist as a permission; Vertex AI uses `aiplatform.endpoints.predict` for invoking predictions on endpoints. Option D is wrong because `aiplatform.models.predict` is a permission for calling prediction directly on a model resource, not on an endpoint, and the error specifically references an endpoint call, not a model call.

114
MCQeasy

A retail company wants to build an assistant that can answer customer questions about its product catalog, but the catalog changes daily. The team wants the model to reference the latest product information without retraining. Which approach should they use?

A.Increase the model's temperature to encourage more creative answers.
B.Fine-tune a foundation model on the product catalog each day.
C.Use retrieval-augmented generation (RAG) to fetch relevant product information at inference time.
D.Deploy the model on a larger machine type with more memory.
AnswerC

RAG retrieves up-to-date documents from an external source and provides them to the model as context, so the assistant can answer based on the latest catalog without retraining. This is ideal when data changes frequently. It also reduces hallucinations because the model grounds its response in retrieved facts.

Why this answer

Retrieval-augmented generation combines a retrieval system with a generative model, allowing the assistant to fetch the latest product details from a database or document store and include them in the prompt. This keeps answers accurate and current without the cost and delay of retraining. It also provides traceability, as the retrieved sources can be cited.

Exam trap

The trap here is assuming that fine-tuning is the only way to inject domain knowledge, when retrieval-augmented generation is specifically designed for frequently changing data.

115
MCQmedium

A healthcare company is using Vertex AI to build a generative AI assistant that helps doctors draft clinical notes. The assistant uses a fine-tuned PaLM 2 model deployed on a private endpoint. Recently, doctors have reported that the assistant takes over 30 seconds to respond, causing workflow delays. Additionally, the monthly Vertex AI costs have increased by 40% without a proportional increase in usage. The model responses are generally accurate but sometimes include irrelevant details. The company wants to improve response time and cost while maintaining acceptable quality. A review of logs shows that most requests are for similar note types (e.g., progress notes, discharge summaries) and that the same prompt is used repeatedly with minor variations. What should the company do first?

A.Switch to a larger model (e.g., Gemini 1.5 Pro) to improve response quality and reduce irrelevant details
B.Increase the Vertex AI endpoint's maximum request quota to handle concurrent requests
C.Apply model quantization (e.g., INT8) to reduce model size and inference time
D.Implement response caching for common queries and batch process similar requests
AnswerD

Caching repeated prompts and batching similar note requests cuts redundant model invocations, directly addressing the 30-second latency and 40% cost increase while preserving output quality. The logs confirm most requests share prompt patterns, making this the highest-impact first step.

Why this answer

The logs show that most requests are for similar note types with repeated prompts, making response caching ideal for reducing latency and cost. Caching stores responses for identical or near-identical queries, eliminating redundant inference calls, which directly addresses the 30-second response time and 40% cost increase without sacrificing quality.

Exam trap

The trap here is that candidates confuse performance optimization techniques (quantization, quota increases) with the root cause of redundant requests, leading them to pick options that address symptoms rather than the fundamental pattern of repeated prompts.

How to eliminate wrong answers

Option A is wrong because switching to a larger model (e.g., Gemini 1.5 Pro) would increase inference time and cost, worsening the problem, and the issue is not about quality but latency and cost. Option B is wrong because increasing the endpoint's maximum request quota does not reduce per-request latency or cost; it only allows more concurrent requests, which could further degrade performance under load. Option C is wrong because model quantization (e.g., INT8) reduces model size and inference time but requires retraining or calibration, and it may degrade accuracy for clinical notes where precision is critical; moreover, it does not address the root cause of redundant requests.

116
MCQmedium

A healthcare company is building a clinical decision support system using Gemini 1.5 Pro on Vertex AI. They need responses that are highly accurate and comply with medical regulations, including traceability to source documents. They have a large corpus of curated medical guidelines stored in PDFs in Cloud Storage. Their team has experience with both fine-tuning and prompt engineering. Which approach best ensures regulatory compliance and accuracy?

A.Use a combination of grounding to the medical guidelines and prompt engineering with system instructions specifying compliance requirements.
B.Use prompt engineering with system instructions and few-shot examples, but no grounding.
C.Use grounding to the medical guidelines but rely on prompt engineering only for compliance instructions.
D.Fine-tune the model on the medical guidelines corpus to internalize the knowledge.
AnswerA

Grounding via Vertex AI Search retrieves passages from the curated PDF guidelines, so every response cites verifiable source documents — satisfying the traceability mandate. System instructions then enforce regulatory constraints at inference time, and because the corpus stays authoritative, accuracy improves without retraining risk. Fine-tuning cannot provide citation-level provenance.

Why this answer

Grounding the model to the curated medical guidelines in Cloud Storage ensures responses are directly traceable to source documents, which is critical for medical regulatory compliance. Combining this with system instructions that specify compliance requirements (e.g., HIPAA, FDA guidelines) enforces behavioral constraints without altering the model's weights, maintaining accuracy and auditability.

Exam trap

The Generative AI Leader exam often tests the misconception that fine-tuning is the best way to ensure accuracy and compliance for domain-specific tasks, but the trap here is that fine-tuning sacrifices traceability and can introduce staleness, whereas grounding with system instructions preserves source attribution and regulatory compliance.

How to eliminate wrong answers

Option B is wrong because relying solely on prompt engineering without grounding provides no mechanism to enforce traceability to specific source documents, making it impossible to meet medical regulatory requirements for evidence-based responses. Option C is wrong because while grounding provides source traceability, relying on prompt engineering alone for compliance instructions is insufficient; system instructions must be explicitly set in the model configuration to ensure consistent enforcement of regulatory constraints across all interactions. Option D is wrong because fine-tuning the model on the medical guidelines corpus internalizes knowledge into the model weights, which can lead to hallucination or outdated information over time, and critically, it breaks traceability to specific source documents since the model cannot cite exact PDF locations or versions.

117
MCQmedium

A financial services company is deploying a generative AI model to summarize sensitive customer emails. The security team requires that no email content is used to train or improve the underlying foundation model. Which Google Cloud approach ensures this requirement is met?

A.Rely on the default data processing terms of Google Cloud, which state that customer data is never used to train foundation models.
B.Fine-tune the model on the email data using Vertex AI, then delete the fine-tuned model after use.
C.Deploy the model in a private VPC and use Cloud VPN to encrypt all traffic between the application and the model endpoint.
D.Use Vertex AI with customer-managed encryption keys (CMEK) and enable VPC Service Controls.
AnswerA

Google Cloud's terms of service for generative AI explicitly state that customer data is not used to train or improve foundation models without explicit permission. This contractual guarantee, combined with technical measures like data residency, ensures the security team's requirement is met. It is the foundational assurance that no email content will be used for training.

Why this answer

Google Cloud's generative AI services are governed by terms that prohibit using customer data to train or improve foundation models without explicit consent. This contractual commitment is the primary control that satisfies the security team's requirement. Technical measures like encryption or network isolation are important for security but do not address the specific concern about training data usage.

Exam trap

The trap here is assuming that network or encryption controls prevent data from being used for model training, when the guarantee is actually contractual and policy-based.

118
MCQeasy

A junior developer is experimenting with a large language model and notices that the same prompt produces different outputs each time it is run. Which characteristic of generative AI models explains this behavior?

A.The model retrains itself after every inference, incorporating the previous output into its weights.
B.The model caches previous responses and randomly selects one from its memory.
C.The model's temperature parameter is fixed at zero, causing deterministic outputs.
D.The model uses a stochastic sampling process to select the next token from a probability distribution.
AnswerD

Generative models assign probabilities to possible next tokens and sample from that distribution. Because sampling is random, repeated runs with the same prompt can yield different outputs. This stochasticity is intentional and controlled by parameters like temperature and top-k. The developer's observation directly reflects this probabilistic, non-deterministic generation process.

Why this answer

The variability in outputs arises because generative models sample from a probability distribution over possible next tokens. This stochastic sampling, influenced by parameters like temperature, leads to different sequences even with identical prompts. The other options incorrectly attribute the behavior to retraining, caching, or a fixed temperature of zero.

Exam trap

The trap here is assuming that a model's output should be deterministic like a traditional function, overlooking the probabilistic sampling that defines generative AI.

119
MCQhard

A healthcare company is using a generative AI model to summarize patient records. They are concerned about the model generating incorrect medical information. They want to ensure that the summaries are grounded in the provided records and do not include hallucinations. Which technique should they implement?

A.Increase the temperature to allow the model to explore more possibilities.
B.Reduce the max output tokens to limit the summary length.
C.Fine-tune the model on a large corpus of medical textbooks.
D.Use retrieval-augmented generation (RAG) to fetch relevant patient data and include it in the prompt.
AnswerD

RAG combines a retrieval system with the generative model to fetch relevant documents (e.g., patient records) and include them in the prompt. This grounds the model's output in actual data, reducing hallucinations. For summarizing patient records, RAG ensures the model references the provided information rather than relying solely on its parametric knowledge.

Why this answer

Retrieval-augmented generation (RAG) is specifically designed to ground model outputs in external knowledge sources. By retrieving relevant patient records and including them in the prompt, the model can generate summaries that are based on the actual data, significantly reducing the risk of hallucinations. This is critical in healthcare where accuracy is paramount.

Exam trap

The trap here is thinking that fine-tuning on medical data will prevent hallucinations in summaries, when in fact grounding requires the model to reference the specific input records, which RAG provides.

120
MCQeasy

A retail company wants to build an internal tool that generates short product descriptions from bullet points. They have no machine learning engineers on staff and want to avoid managing any infrastructure. Which Google Cloud approach best fits their need?

A.Use BigQuery ML to create a linear regression model that predicts description length.
B.Train a custom image generation model on Vertex AI using their product photos.
C.Deploy an open-source LLM on a Compute Engine VM with a GPU and expose it via a REST API.
D.Use the Gemini API in Vertex AI to generate descriptions from the bullet points with a prompt.
AnswerD

The Gemini API in Vertex AI provides a fully managed, serverless way to call a foundation model for text generation. The team only needs to craft a prompt containing the bullet points and receive generated descriptions, with no infrastructure or ML engineering required. This directly matches the scenario's need for quick, low-effort text generation.

Why this answer

The Gemini API in Vertex AI is a managed service that lets teams generate text without building or operating ML infrastructure. By supplying bullet points in a prompt, the model produces product descriptions on demand. This aligns with the company's lack of ML engineers and desire to avoid infrastructure management, making it the most practical and direct solution.

Exam trap

The trap here is assuming that any Google Cloud AI service can generate text, when many services like BigQuery ML or custom training are designed for predictive or image tasks, not generative text.

121
MCQmedium

A data scientist notices that a Gemini model generates inconsistent responses to similar prompts. What is the likely cause?

A.Model is not fine-tuned enough
B.The prompt is too short
C.The temperature setting is too low
D.The top_p or temperature parameters are set too high causing randomness
AnswerD

Temperature and top_p control sampling randomness; high values widen the probability distribution over candidate tokens. That directly produces the inconsistent outputs observed for similar prompts, since the model samples differently each call rather than selecting the highest-probability token.

Why this answer

High temperature (e.g., >1.0) or high top_p (e.g., >0.9) increases the randomness of token sampling, causing the model to select less probable tokens. This directly leads to inconsistent responses for similar prompts, as the model's output distribution becomes more uniform and less deterministic.

Exam trap

Google Cloud often tests the misconception that fine-tuning or prompt length is the primary cause of output inconsistency, when in fact the sampling parameters (temperature and top_p) directly control randomness and are the most common culprit.

How to eliminate wrong answers

Option A is wrong because fine-tuning adjusts the model's weights for a specific task, but it does not control the randomness of token generation; even a fully fine-tuned model will produce inconsistent outputs if sampling parameters are set too high. Option B is wrong because prompt length affects context and specificity, not the inherent randomness of the generation process; a short prompt can still yield consistent responses if temperature and top_p are low. Option C is wrong because a low temperature setting (e.g., 0.1) actually reduces randomness, making outputs more deterministic and consistent, not inconsistent.

122
Multi-Selectmedium

Which THREE of the following are common techniques to reduce harmful biases in generative AI models? (Choose three.)

Select 3 answers
A.Use reinforcement learning from human feedback (RLHF) with a reward model that penalizes biased or unfair outputs.
B.Curate diverse and balanced training datasets that overrepresent underrepresented groups.
C.Decrease the model's temperature parameter to make outputs more deterministic.
D.Apply adversarial training to remove protected attribute information from hidden representations.
E.Conduct a legal review of all generated outputs before release.
AnswersA, B, D

RLHF directly targets bias by training a reward model that assigns lower scores to unfair outputs, then optimising the generative model against it. This satisfies the stem's requirement for a bias-reduction technique, since the penalty signal actively steers the model away from biased generations during fine-tuning rather than merely detecting them afterwards.

Why this answer

Option A is correct because RLHF with a reward model that explicitly penalizes biased or unfair outputs directly optimizes the generative model's policy toward fairness-aligned behavior, using human preference data to shape the reward signal. Option B is correct because curating diverse and balanced training datasets—deliberately overrepresenting underrepresented groups—counteracts sampling bias and spurious correlations learned from skewed data, which is a foundational bias-mitigation technique. Option D is correct because adversarial training that removes protected attribute information from hidden representations (e.g., via adversarial classifiers or fair representation learning) reduces the model's ability to encode and exploit sensitive attributes.

Option C does not belong because lowering the temperature only makes sampling more deterministic; it changes randomness, not the underlying learned biases, and can even amplify the most probable (biased) output. Option E does not belong because a legal review of outputs is a post-hoc compliance check, not a technique that reduces bias within the generative model itself.

Exam trap

Google Cloud often tests the distinction between hyperparameter tuning (like temperature) and actual bias mitigation techniques, so candidates mistakenly think lowering temperature reduces bias when it only affects output randomness.

123
MCQeasy

A company wants to generate images from text descriptions using Google Cloud. Which service should they use?

A.Vertex AI Imagen
B.Vertex AI Gemini
C.Cloud Vision API
D.AutoML Vision
AnswerA

Vertex AI Imagen generates images directly from text prompts, satisfying the stem's requirement for text-to-image generation on Google Cloud. Unlike Vertex AI's language or vision models, Imagen is purpose-built for photorealistic image synthesis, offering resolution, editing and watermark controls that match this scenario precisely.

Why this answer

Vertex AI Imagen is Google Cloud's purpose-built service for generating high-fidelity images from text descriptions using diffusion models. It directly addresses the requirement of text-to-image generation, offering capabilities like image editing, upscaling, and style transfer, which are not available in other Vertex AI or Vision services.

Exam trap

The trap here is that candidates may confuse Vertex AI Gemini's multimodal capabilities (understanding images) with generative image creation, or assume that Cloud Vision API or AutoML Vision can be repurposed for generation, when in fact they are strictly analysis or custom training tools.

How to eliminate wrong answers

Option B is wrong because Vertex AI Gemini is a multimodal large language model (LLM) that can process text, images, audio, and video, but it is not optimized or primarily designed for generating images from text; its strength lies in understanding and reasoning across modalities, not in image synthesis. Option C is wrong because Cloud Vision API is a pre-trained model for analyzing and extracting information from images (e.g., object detection, OCR, label detection), not for generating images from text. Option D is wrong because AutoML Vision is a service for training custom image classification or object detection models on labeled datasets, not for generative text-to-image tasks.

124
MCQmedium

A retail company wants to build an internal assistant that answers employee questions using the company's HR policy documents. The documents change frequently, and the company does not want to retrain a model every time a policy is updated. Which approach best meets these requirements on Google Cloud?

A.Fine-tune a foundation model on the HR documents each time a policy changes
B.Increase the model's context window and paste all HR documents into every prompt
C.Deploy the model with a lower temperature and rely on its pretrained HR knowledge
D.Use retrieval-augmented generation (RAG) with a vector store of the HR documents
AnswerD

RAG retrieves relevant passages from an indexed vector store at query time and supplies them to the model as context, so updated documents only need re-indexing, not retraining. This satisfies the frequent-change requirement and grounds answers in authoritative sources, reducing hallucination. It is the standard Google Cloud pattern using Vertex AI Search or a vector database with embeddings.

Why this answer

Retrieval-augmented generation separates knowledge from the model: documents are embedded and indexed, and relevant chunks are retrieved and passed as context for each query. Updating a policy means re-indexing that document, not retraining, which directly meets the frequent-change requirement. It also grounds responses in source material, improving accuracy and enabling citations, which is essential for an HR policy assistant.

Exam trap

The trap here is conflating fine-tuning with knowledge injection, when fine-tuning is for behavior and style, not for frequently changing factual content.

125
MCQmedium

A data scientist fine-tunes a large language model on Vertex AI but gets poor results on validation data. What is the most likely cause?

A.Incorrect learning rate
B.Insufficient training data
C.Using wrong model family
D.Overfitting due to too many epochs
AnswerB

Fine-tuning adapts an existing model rather than teaching new knowledge from scratch, so too few examples leave it unable to learn the target pattern. Insufficient training data therefore produces poor validation results, matching the stem's symptom.

Why this answer

Fine-tuning a large language model on Vertex AI with poor validation results is most likely due to insufficient training data. Large language models have billions of parameters and require a substantial amount of high-quality, task-specific data to effectively adapt to a new domain or task; without enough examples, the model cannot learn the desired patterns and will perform poorly on unseen data.

Exam trap

The trap here is that candidates often assume hyperparameter tuning (like learning rate) is the primary cause of poor fine-tuning results, but in generative AI, data quantity and quality are the most common bottlenecks, especially when using pre-trained models on Vertex AI.

How to eliminate wrong answers

Option A is wrong because an incorrect learning rate typically causes training instability (e.g., loss divergence or slow convergence) rather than consistently poor validation results, and Vertex AI's default hyperparameters are often reasonable. Option C is wrong because using the wrong model family (e.g., choosing a text generation model for a classification task) would likely cause immediate, obvious failures or mismatches in output format, not just poor validation performance after fine-tuning. Option D is wrong because overfitting due to too many epochs would manifest as high training accuracy with low validation accuracy, but the question states poor results on validation data without mentioning training performance, and overfitting is less likely with insufficient data (the model would underfit instead).

126
MCQhard

A financial services firm is deploying a generative AI chatbot on Vertex AI. They need to ensure that the model's responses comply with strict regulatory requirements and do not include personally identifiable information (PII). They want to automatically filter out any PII from both user inputs and model outputs. Which Google Cloud service should they integrate?

A.Cloud Data Loss Prevention (DLP) API
B.Security Command Center
C.Cloud Armor
D.Cloud Identity-Aware Proxy (IAP)
AnswerA

The Cloud Data Loss Prevention (DLP) API is designed to discover, classify, and protect sensitive data such as PII. It can inspect text for PII and redact or mask it. By integrating DLP with the chatbot pipeline, the firm can automatically scan and filter user inputs and model outputs, ensuring compliance. This is the appropriate service for PII detection and redaction in text.

Why this answer

To automatically detect and redact PII from text, the Cloud Data Loss Prevention (DLP) API is the correct service. It can be integrated into the chatbot's processing pipeline to scan both incoming user messages and outgoing model responses, identifying and removing sensitive data according to predefined or custom infoTypes. This ensures regulatory compliance and protects user privacy.

Exam trap

The trap here is confusing security services that protect infrastructure with data-level protection services that inspect and redact content.

127
MCQhard

A developer uses Vertex AI to generate code but the output is not syntactically correct. Which parameter should be adjusted?

A.candidate_count
B.max_output_tokens
C.temperature
D.top_k
AnswerC

Temperature controls sampling randomness: higher values increase diversity but raise the chance of malformed syntax, while lower values make token selection more deterministic. Reducing temperature steers Vertex AI toward the most probable, syntactically valid continuations, directly addressing the incorrect code output.

Why this answer

Temperature controls the randomness of token selection during generation. A high temperature increases the likelihood of less probable tokens, which can lead to syntactically incorrect code. Lowering temperature makes the model more deterministic and conservative, favoring higher-probability tokens that are more likely to form valid syntax.

Exam trap

Google Cloud often tests the misconception that increasing candidate_count or max_output_tokens will improve output quality, when in fact these parameters only affect quantity or length, not the underlying token selection logic that determines syntactic correctness.

How to eliminate wrong answers

Option A is wrong because candidate_count controls how many different response candidates are generated, not the syntactic correctness of any single output. Option B is wrong because max_output_tokens limits the length of the generated text, not the quality or validity of the syntax. Option D is wrong because top_k limits the number of highest-probability tokens considered at each step; while it can affect output quality, it does not directly address syntactic correctness as effectively as temperature does.

128
MCQmedium

A company wants to use a pre-trained language model for customer support summarization. They need to ensure responses are concise and accurate. Which prompt engineering technique is most effective?

A.Zero-shot prompting
B.Few-shot prompting with examples
C.Chain-of-thought prompting
D.Negative prompting
AnswerB

Few-shot prompting supplies labelled input-output pairs, so the model infers the required summary length and factual style directly from the examples. This steers formatting and conciseness without retraining, satisfying the demand for concise, accurate customer support summaries.

Why this answer

Few-shot prompting (B) is most effective because it provides the model with a small set of example input-output pairs (e.g., a customer query and its concise summary), which guides the model to produce outputs that match the desired format, length, and accuracy. This technique is particularly useful for summarization tasks where consistency and adherence to a specific style are critical, as it reduces ambiguity without requiring fine-tuning.

Exam trap

Google Cloud often tests the misconception that zero-shot prompting is sufficient for all tasks, but the trap here is that candidates overlook the need for explicit guidance in format-sensitive tasks like summarization, where few-shot examples provide the necessary constraint for consistency.

How to eliminate wrong answers

Option A (Zero-shot prompting) is wrong because it relies solely on the model's pre-trained knowledge without any examples, which often leads to inconsistent or overly verbose summaries, especially when the task requires a specific format or level of conciseness. Option C (Chain-of-thought prompting) is wrong because it is designed for multi-step reasoning tasks (e.g., arithmetic or logic problems) and is unnecessary for summarization, where the goal is to condense information rather than reason through steps. Option D (Negative prompting) is wrong because it instructs the model on what to avoid (e.g., 'do not include details'), which can be imprecise and may inadvertently suppress relevant information, making it less reliable than providing positive examples of desired outputs.

129
MCQhard

A financial services firm is deploying a generative AI model to answer customer queries about investment products. They want to ensure the model's responses are based on the most current and authoritative internal documents. Which technique should they implement?

A.Using retrieval-augmented generation (RAG) with a vector database
B.Deploying the model with a larger context window
C.Fine-tuning the model on historical customer interactions
D.Increasing the model's temperature parameter
AnswerA

RAG combines a generative model with a retrieval system that fetches relevant documents from a vector database in real time. By indexing the firm's current authoritative documents, the model can ground its responses in the latest information. This ensures accuracy and reduces hallucinations, directly meeting the requirement for responses based on current internal documents.

Why this answer

Retrieval-augmented generation (RAG) dynamically retrieves relevant documents from a vector database and provides them to the model as context, ensuring responses are grounded in the most current authoritative sources. This is ideal for the financial firm's need to answer queries based on up-to-date internal documents. Other options do not provide real-time grounding.

Exam trap

The trap here is thinking that fine-tuning or a larger context window alone can keep responses current, when they lack dynamic retrieval.

130
MCQmedium

A marketing team is using a generative AI model to create ad copy. They want to ensure the output is creative and diverse, but also relevant to the product. They decide to adjust the model's parameters. Which parameter should they increase to make the output more diverse?

A.Top-k
B.Temperature
C.Top-p
D.Max output tokens
AnswerB

Temperature controls the randomness of the model's predictions. A higher temperature increases diversity by making the probability distribution flatter, so less likely words are chosen more often. This leads to more creative and varied outputs. For ad copy, increasing temperature can help generate more diverse ideas while still being relevant if the prompt is clear.

Why this answer

Temperature is the parameter that directly controls the randomness of the model's output. Increasing temperature makes the output more diverse and creative, which is desired for ad copy generation. While top-k and top-p also influence sampling, temperature is the primary knob for adjusting diversity.

Max output tokens only controls length.

Exam trap

The trap here is confusing parameters that control output length or sampling cutoffs with the one that directly scales randomness, which is temperature.

131
MCQmedium

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?

A.Increase the batch size to stabilize training.
B.Decrease the number of fine-tuning steps to prevent overfitting.
C.Include examples of informal customer interactions in the fine-tuning data.
D.Use a higher learning rate for faster adaptation.
AnswerC

Fine-tuning adapts a model's style to its training distribution, so the formal tone reflects the transcripts used. Adding informal customer interaction examples shifts that distribution toward conversational phrasing, directly satisfying the requirement to make responses less formal without changing architecture or decoding parameters.

Why this answer

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.

132
MCQeasy

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

A.AI Platform Notebooks
B.Dialogflow CX
C.Vertex AI Generative AI Studio
D.Cloud Composer
AnswerC

Vertex AI Generative AI Studio supplies a managed workspace for prompt design, tuning and model evaluation, directly satisfying the stem's requirement for a managed prompt engineering and evaluation environment. Its integrated prompt gallery and evaluation metrics remove infrastructure overhead, unlike raw Vertex AI endpoints or unmanaged notebook approaches.

Why this answer

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.

Exam trap

The trap here is that candidates may confuse Vertex AI Generative AI Studio with AI Platform Notebooks, assuming that any model development environment supports prompt engineering, when in fact Generative AI Studio is the specialized tool for that purpose.

How to eliminate wrong answers

Option A is wrong because AI Platform Notebooks is a managed Jupyter notebook service for custom model development and training, not a dedicated environment for prompt engineering or model evaluation. Option B is wrong because Dialogflow CX is a conversational AI platform for building chatbots and virtual agents, focused on intent classification and dialogue management, not prompt engineering or model evaluation. Option D is wrong because Cloud Composer is a managed Apache Airflow service for workflow orchestration and scheduling, unrelated to prompt engineering or model evaluation.

133
Multi-Selecthard

A healthcare company is evaluating foundation models for a patient triage assistant. They need to ensure the model is appropriate for medical use and that they can control costs. Which two factors should they consider when selecting and using a foundation model in Vertex AI? (Choose two.)

Select 2 answers
A.The model's ability to generate images for patient education materials.
B.The color scheme of the model's documentation website.
C.The token-based pricing and expected request volume to estimate and control costs.
D.The model's training data cutoff and domain relevance to medical knowledge.
E.The number of GPUs available in the company's on-premises data center.
AnswersC, D

Generative AI usage in Vertex AI is typically priced by tokens processed. Understanding token-based pricing and forecasting request volume allows the company to estimate costs and apply controls like quotas or budget alerts. This is essential for managing spend in a patient triage assistant that may handle many interactions, making it a key selection and usage factor.

Why this answer

For a medical triage assistant, the model's training data cutoff and domain relevance are critical to ensure safe, current guidance. Token-based pricing and expected volume determine cost predictability and control. Together, these factors address the dual requirements of clinical appropriateness and financial management when selecting and using a foundation model in Vertex AI.

Exam trap

The trap here is assuming that infrastructure details like on-premises GPUs matter for a managed service, or that unrelated features like image generation are relevant to a text triage assistant.

134
MCQeasy

A developer is using a generative AI model to summarize long articles. They want to ensure the summaries are concise and do not exceed a certain length. Which parameter should they adjust to control the maximum length of the generated summary?

A.Temperature
B.Max output tokens
C.Top-p
D.Top-k
AnswerB

Max output tokens sets the maximum number of tokens the model can generate in its response. By lowering this value, the developer can ensure summaries stay within a desired length. This parameter directly caps the output size, making it the correct choice for controlling summary length.

Why this answer

The max output tokens parameter directly limits the number of tokens the model can generate. By setting it to a desired value, the developer can prevent summaries from becoming too long. Other parameters like temperature, top-p, and top-k influence randomness and diversity but do not control output length.

Exam trap

The trap here is confusing parameters that control randomness with those that control output length, such as assuming temperature affects length.

135
Multi-Selectmedium

A company is deploying a generative AI model for medical diagnosis support. Which THREE considerations are critical for responsible AI?

Select 3 answers
A.Ensure the training data is diverse and representative.
B.Maximize model throughput to handle high volumes.
C.Implement human oversight for all diagnostic suggestions.
D.Provide clear disclaimers about the model's limitations.
E.Use the cheapest model to reduce costs.
AnswersA, C, D

Diverse, representative training data reduces demographic bias, preventing skewed diagnostic suggestions for under-represented patient groups. This satisfies the responsible AI requirement by addressing fairness at the data layer, where bias originates before model training begins.

Why this answer

Option A is correct because diverse and representative training data is essential to reduce bias and ensure the model performs equitably across different patient demographics, which is a core responsible AI principle in medical contexts. Option C is correct because human oversight for all diagnostic suggestions ensures that qualified clinicians review and validate AI outputs, maintaining accountability and preventing harm from erroneous predictions. Option D is correct because clear disclaimers about the model's limitations inform users of its scope and uncertainty, preventing over-reliance and supporting informed decision-making.

Option B is not correct because maximizing throughput addresses performance and scalability, not responsible AI concerns such as fairness, safety, or transparency. Option E is not correct because choosing the cheapest model prioritizes cost over accuracy, safety, and ethical considerations, which is inappropriate for medical diagnosis support.

Exam trap

Google Cloud often tests the distinction between operational metrics (like throughput or cost) and ethical/regulatory requirements (like fairness, transparency, and human oversight) in responsible AI, leading candidates to mistakenly select performance-based options as critical considerations.

136
MCQeasy

What is the purpose of grounding in Vertex AI?

A.To improve training speed
B.To connect model outputs to verifiable sources
C.To reduce model size for faster inference
D.To enable multi-modal inputs
AnswerB

Grounding links model outputs to verifiable sources such as search results or enterprise data, reducing hallucination. This satisfies the need to trace generated claims back to authoritative evidence, ensuring responses are factually anchored rather than relying solely on parametric knowledge.

Why this answer

Grounding in Vertex AI connects model outputs to verifiable, external sources of information (such as Google Search, enterprise data sources, or third-party databases) to reduce hallucinations and improve factual accuracy. By referencing grounded sources, the model can provide citations and allow users to verify claims, which is critical for enterprise applications requiring trust and compliance.

Exam trap

Google Cloud often tests grounding by conflating it with fine-tuning or prompt engineering, so the trap here is assuming grounding modifies the model's weights or training process, when in fact it is a retrieval-based augmentation layer applied at inference time.

How to eliminate wrong answers

Option A is wrong because grounding does not improve training speed; it is a runtime technique applied during inference to augment responses with real-time data, not a training optimization. Option C is wrong because grounding does not reduce model size or accelerate inference; it may actually add latency due to the retrieval step. Option D is wrong because grounding is not about enabling multi-modal inputs; it specifically addresses output verification and source attribution, whereas multi-modal support is a separate capability for processing images, audio, or video alongside text.

137
MCQmedium

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?

A.Add more training data
B.Increase the learning rate
C.Apply dropout
D.Use a larger batch size
AnswerC

Dropout randomly deactivates units during training, forcing the network to learn redundant, distributed representations instead of memorising training samples. This regularisation narrows the gap between training and validation performance, directly addressing the poor generalisation the team observed during model evaluation.

Why this answer

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.

Exam trap

Google Cloud often tests the distinction between techniques that improve generalization (regularization) versus those that improve optimization (learning rate, batch size), leading candidates to confuse data augmentation or hyperparameter tuning with regularization methods like dropout.

How to eliminate wrong answers

Option A is wrong because adding more training data can help reduce overfitting but is not a regularization technique; it addresses data scarcity, not the core issue of model complexity. Option B is wrong because increasing the learning rate can cause training instability, divergence, or overshooting of the loss minimum, and does not prevent overfitting. Option D is wrong because using a larger batch size often leads to sharper minima and poorer generalization, potentially worsening overfitting, and is not a regularization method.

138
MCQeasy

A marketing team wants to generate social media posts with Google Cloud's generative AI, but they have no machine learning experience. They need a no-code interface to experiment with prompts and tune settings like temperature. Which Google Cloud service should they use?

A.Document AI
B.BigQuery ML
C.Dialogflow CX
D.Vertex AI Studio
AnswerD

Vertex AI Studio provides a visual, no-code console where users can design prompts, choose models like Gemini, adjust parameters such as temperature and top-k, and immediately view generated text. It is designed for rapid experimentation without writing code, making it ideal for the marketing team's need to prototype social media content.

Why this answer

Vertex AI Studio is the correct choice because it offers a no-code environment specifically for generative AI experimentation. It allows users to write prompts, select foundation models, and adjust generation parameters, then see results instantly. This matches the marketing team's need to create social media posts without ML expertise.

Exam trap

The trap here is confusing Vertex AI Studio with other Google Cloud AI services that are not designed for no-code generative text experimentation.

139
MCQeasy

A retail company wants to use a generative AI model to create product descriptions from a list of attributes. They want a fully managed service that allows them to quickly experiment with prompts and adjust parameters without managing infrastructure. Which Google Cloud service should they use?

A.AutoML Natural Language
B.Cloud Natural Language API
C.Dialogflow CX
D.Vertex AI Studio
AnswerD

Vertex AI Studio is a Google Cloud console tool that lets you design, test, and manage prompts for generative AI models. It provides an interactive interface to experiment with different models and parameters, such as temperature and top-k, without managing any infrastructure. This directly matches the requirement for quick experimentation with prompts and parameters in a fully managed environment.

Why this answer

Vertex AI Studio is a Google Cloud service that provides a unified interface for experimenting with generative AI models. It allows users to write prompts, adjust model parameters, and see generated outputs in real time. Since the company wants a fully managed service to quickly create product descriptions without infrastructure management, Vertex AI Studio is the correct choice.

Exam trap

The trap here is confusing generative AI tools with traditional NLP services that analyze rather than generate text.

140
MCQhard

A healthcare analytics team is evaluating a generative AI model for summarizing clinical notes. They observe that summaries are fluent but occasionally invent details not present in the source notes. They want a practical mitigation that reduces fabricated content without retraining the model. Which approach should they apply first?

A.Raise the temperature to encourage the model to explore alternative phrasings of the notes
B.Fine-tune the model on a large corpus of unrelated medical literature to improve fluency
C.Lower the temperature and instruct the model to answer only from the provided note text
D.Increase the maximum output token limit so the model has more room to explain itself
AnswerC

Lowering temperature reduces random sampling, and an explicit instruction to use only the supplied note constrains the model to the given context. Together they reduce the chance of invented details while keeping the model unchanged. This is a low-cost, immediate mitigation that targets the observed hallucination without retraining.

Why this answer

Reducing temperature and explicitly instructing the model to rely only on the provided note constrains generation toward the source content, cutting down fabricated details without retraining. It is fast, inexpensive, and directly targets the failure mode. Longer output, higher randomness, or unrelated fine-tuning do not improve fidelity and can make hallucination worse.

Exam trap

The trap here is assuming more output space or more training data fixes hallucination, when the first effective lever is constraining sampling and instructing the model to use only the supplied context.

141
MCQmedium

A healthcare company is using a generative AI model to summarize patient notes. The model occasionally includes hallucinated medical details. Which strategy best reduces hallucinations?

A.Use a larger model with more parameters.
B.Fine-tune the model on a large corpus of general medical literature.
C.Increase the model's temperature to encourage more diverse outputs.
D.Provide the model with the patient notes as context and instruct it to only use that information.
AnswerD

Grounding the model in the source text and explicitly instructing it to rely only on that context reduces hallucinations because the model can copy or paraphrase from the provided notes. This is a form of retrieval-augmented generation or prompt grounding. It constrains the model's output to the given facts.

Why this answer

Grounding the model by providing the patient notes as context and instructing it to only use that information is the most effective way to reduce hallucinations. This approach, often called retrieval-augmented generation or prompt grounding, forces the model to base its summary on the given text rather than relying on its parametric memory. It also allows for verification against the source.

Exam trap

The trap here is assuming that a larger or fine-tuned model will automatically be more factual, when the key is to constrain generation to the provided source text.

142
MCQeasy

You are a generative AI lead at a healthcare startup developing a system to summarize patient medical records for quick review by doctors. The system uses a fine-tuned LLM. After deployment, doctors report that the summaries often miss critical details like medication dosages and allergy information. The current pipeline preprocesses patient records by extracting text from EHR, feeding it to the LLM, and outputting a summary. The team has limited time and budget. They cannot retrain the model because it is hosted as a managed API. Which action should you take to most effectively improve the summarization quality without changing the model?

A.Increase the maximum output token limit to force the model to include more details.
B.Replace the LLM with a simpler extractive summarization model that selects sentences from the original document.
C.Implement a retrieval-augmented generation (RAG) system that pulls supplementary data from external drug databases.
D.Revise the prompt to explicitly ask for medication dosages and allergies, and format the input text by adding headings (e.g., '### Medications') to emphasize important sections.
AnswerD

Prompt engineering and structured input formatting directly address the omission of dosages and allergies without touching the managed API model, satisfying the no-retraining constraint. Explicit instructions plus section headings steer the LLM's attention to clinically critical fields, a low-cost, rapid fix suited to the limited time and budget.

Why this answer

Prompt engineering is the most effective and cost-efficient way to improve LLM output without retraining or changing the model. By explicitly instructing the model to include medication dosages and allergies, and by structuring the input with clear headings, you guide the model's attention to critical sections, directly addressing the missing details. This approach leverages the LLM's existing capabilities and requires no changes to the hosted API or additional infrastructure.

Exam trap

Google Cloud often tests the misconception that increasing output length or adding external data automatically improves quality, when in fact the most direct and cost-effective fix is to refine the input prompt to guide the model's focus.

How to eliminate wrong answers

Option A is wrong because increasing the maximum output token limit does not force the model to include specific missing details; it only allows longer responses, which may still omit critical information if the prompt does not direct the model's focus. Option B is wrong because replacing the LLM with an extractive summarization model would require retraining or deploying a new model, contradicting the constraint of not changing the model, and extractive methods cannot generate new text to explicitly mention dosages or allergies if they are not present in the original text. Option C is wrong because implementing a RAG system to pull from external drug databases adds complexity, cost, and latency, and does not address the core issue of missing details from the patient's own records; the problem is about extracting existing information, not supplementing with external data.

143
Multi-Selecthard

A company is fine-tuning a Gemma model using Vertex AI. They observe that the model overfits. Which TWO actions should they take to mitigate overfitting?

Select 2 answers
A.Use a larger batch size
B.Increase the number of training epochs
C.Use more diverse data
D.Reduce the learning rate
E.Add dropout during fine-tuning
AnswersC, E

Overfitting arises when training data is too narrow, letting the model memorise specifics rather than general patterns. Expanding the fine-tuning dataset with more varied examples exposes Gemma to broader linguistic patterns, improving generalisation and reducing the gap between training and validation performance.

Why this answer

Option C (Use more diverse data) is correct because overfitting occurs when the fine-tuning dataset is too narrow or unrepresentative, causing the model to memorize specific patterns rather than generalize; expanding the training data with more varied examples exposes the model to a broader distribution and reduces memorization. Option E (Add dropout during fine-tuning) is correct because dropout randomly deactivates a fraction of neurons during training, which acts as a regularization technique that prevents the model from relying too heavily on any single pathway and improves generalization. Option A (Use a larger batch size) is not a reliable overfitting remedy and can even reduce regularization effects from gradient noise, so it is not among the correct answers.

Option B (Increase the number of training epochs) would worsen overfitting by allowing the model to fit the training data even more closely. Option D (Reduce the learning rate) may stabilize training but does not directly address overfitting and can, in some cases, allow the model to converge to a sharper minimum that generalizes poorly.

Exam trap

Google Cloud often tests the misconception that reducing the learning rate or increasing batch size are universal fixes for overfitting, when in fact these hyperparameters primarily affect optimization dynamics rather than regularization.

144
MCQeasy

A financial analyst wants to quickly generate a first draft of a market commentary from a few key statistics using a Google Cloud generative AI model. They have no coding experience and want to see the model's response immediately while adjusting the prompt. What should they do first?

A.Build a pipeline in Vertex AI Pipelines to preprocess the statistics before generation.
B.Open Vertex AI Studio, select a generative model, and enter the statistics and instructions in the prompt box.
C.Create a custom notebook in Colab Enterprise and install the necessary libraries before prompting.
D.Write a Python script using the Vertex AI SDK to call the model and print the output.
AnswerB

Vertex AI Studio provides a no-code interface where users select a model, type a prompt, and immediately view generated text. It is designed for exactly this kind of rapid drafting and prompt iteration. The analyst can adjust wording and parameters interactively, making it the fastest and most appropriate first step without writing code.

Why this answer

Vertex AI Studio is the Google Cloud tool for interactive, code-free prompt experimentation. An analyst can choose a model, type statistics and instructions, and read the generated commentary right away. This matches the need for a quick first draft and immediate prompt adjustments better than any code-based or pipeline-based approach.

Exam trap

The trap here is reaching for code or pipelines because they are powerful, when the requirement is simply fast, no-code interactive generation.

145
MCQmedium

A marketing team is using a generative AI model to create ad copy. They want to control the model's creativity so that outputs are more focused and deterministic for a formal campaign. Which parameter should they adjust?

A.Temperature
B.Max output tokens
C.Top-p (nucleus sampling)
D.Top-k sampling
AnswerA

Temperature directly controls the randomness of the model's output. Lowering the temperature makes the model more deterministic and focused, which is ideal for a formal campaign where consistency is key. It reduces the likelihood of creative but off-brand outputs, aligning with the need for controlled creativity.

Why this answer

Temperature is the key parameter for controlling randomness; lowering it makes output more deterministic and focused. This helps ensure ad copy is consistent and on-brand for a formal campaign. Other parameters affect length or token selection but do not directly control creativity in the same way.

Exam trap

The trap here is confusing parameters that affect diversity (like top-p or top-k) with the primary control for determinism, which is temperature.

146
MCQeasy

A startup wants to use a pre-trained model to generate product descriptions without training. Which Google Cloud service should they use?

A.Vertex AI Prediction
B.AI Platform Training
C.Cloud AutoML
D.Vertex AI Generative AI Studio
AnswerD

Vertex AI Generative AI Studio provides access to pre-trained foundation models with prompt design and tuning tools, letting the startup generate product descriptions through prompting alone. No custom training is required, satisfying the constraint of using a pre-trained model without training.

Why this answer

Vertex AI Generative AI Studio is the correct service because it provides a no-code interface to access and experiment with pre-trained generative models, including text generation for product descriptions, without requiring any training or custom model development. It allows users to directly prompt models like PaLM 2 or Gemini for inference tasks, making it ideal for generating content from a pre-trained model without training.

Exam trap

The trap here is that candidates may confuse Vertex AI Prediction (which serves custom models) with Generative AI Studio (which serves pre-trained models), or assume that any generative AI task requires training via AI Platform Training or AutoML, when in fact the question explicitly states 'without training'.

How to eliminate wrong answers

Option A is wrong because Vertex AI Prediction is used for deploying and serving custom-trained models for online or batch predictions, not for directly accessing pre-trained generative models without training. Option B is wrong because AI Platform Training is designed for training custom machine learning models, not for using pre-trained models for inference without training. Option C is wrong because Cloud AutoML is used for training custom models on user-provided data with automated machine learning, not for directly generating content from a pre-trained model without any training.

147
MCQmedium

A developer runs this command: `gcloud ai models upload --region=us-central1 --display-name=my-model --artifact-uri=gs://my-bucket/model.pkl`. What is the primary purpose?

A.Create a training pipeline
B.Deploy a model to an endpoint
C.Train a model
D.Upload a model artifact to Model Registry
AnswerD

The command registers a trained model artefact stored in Cloud Storage with Vertex AI Model Registry in us-central1, assigning the display name my-model. It does not deploy an endpoint or run training; it creates the registry entry for versioning and deployment.

Why this answer

The command `gcloud ai models upload` uploads a local model artifact (model.pkl stored in Cloud Storage) to the Vertex AI Model Registry. This is used for versioning and managing trained models, not for initiating training, deployment, or pipeline creation. The Model Registry stores the model artifact for later use in deployments or predictions.

Exam trap

Google Cloud exams often test the distinction between model registration (uploading a trained artifact) and model training or deployment. Candidates may confuse `gcloud ai models upload` with starting a training job or deploying a model, but this command only stores the model artifact in the registry for versioning and reuse.

How to eliminate wrong answers

Option A is wrong because creating a training pipeline requires a command like `az ml job create` or `az ml pipeline create`, not `az ml model create`. Option B is wrong because deploying a model to an endpoint uses commands such as `az ml online-endpoint create` and `az ml online-deployment create`, which involve specifying compute targets and scoring scripts, not just uploading a model file. Option C is wrong because training a model is performed via a training job (e.g., `az ml job create` with a training script and compute target), not by registering an already-trained artifact.

148
MCQeasy

A company is deploying a generative AI model for medical advice. What is the most important consideration?

A.Model latency
B.Safety and fairness
C.Model size
D.Cost of inference
AnswerB

Medical advice carries direct patient-safety and regulatory risk, so outputs must avoid harmful or biased recommendations. Safety and fairness filters and evaluation address this constraint, ensuring the generative model does not produce unsafe guidance or discriminate across patient groups.

Why this answer

In medical advice applications, a generative AI model's outputs can directly impact patient health, making safety and fairness the paramount consideration. Incorrect or biased advice could lead to misdiagnosis or harm, outweighing performance metrics like latency or cost. Regulatory frameworks such as HIPAA and FDA guidelines for clinical decision support further mandate rigorous validation of model safety and fairness before deployment.

Exam trap

Google Cloud often tests the misconception that technical performance metrics like latency or cost are the primary concerns in high-stakes domains, when in fact ethical and safety considerations take precedence.

How to eliminate wrong answers

Option A is wrong because model latency, while important for user experience, is secondary to ensuring the advice is safe and unbiased; a fast but harmful response is unacceptable in healthcare. Option C is wrong because model size correlates with computational resources and potential capability, but does not inherently guarantee safety or fairness; a larger model may amplify biases or generate more confident but incorrect advice. Option D is wrong because cost of inference is a business consideration that must be balanced against safety requirements, but it is not the most critical factor when human lives are at stake.

149
MCQmedium

A developer is building a customer support chatbot using Gemini models on Vertex AI. The chatbot must respond in a consistent, professional tone and avoid generating harmful or off-topic content. Which approach should the developer take to enforce these behavioral constraints?

A.Use a system instruction to define the chatbot's role and guidelines.
B.Increase the top-k parameter to 40.
C.Enable the model's safety filters to block all categories.
D.Set the temperature parameter to 0.
AnswerA

System instructions are a feature in Vertex AI that allow developers to set the model's behavior, tone, and constraints at the start of a conversation. By providing a system instruction that specifies a professional tone and prohibits harmful content, the developer can guide the model's responses consistently across all interactions.

Why this answer

System instructions are the correct mechanism to set persistent behavioral guidelines for a generative AI model. They are processed before user input and apply to the entire conversation, ensuring the chatbot maintains a professional tone and avoids harmful or off-topic content. Other parameters like temperature or top-k affect creativity, not adherence to rules.

Exam trap

The trap here is assuming that safety filters or low temperature can replace explicit behavioral instructions for tone and topic control.

150
MCQhard

A media company is using a large language model to generate scripts for animated shorts. They notice that the model sometimes produces content that closely resembles existing copyrighted scripts from well-known movies. Which technique should they implement to reduce the risk of generating such content?

A.Use a filtered decoding strategy that blocks outputs matching known copyrighted content.
B.Apply a repetition penalty to discourage the model from repeating phrases.
C.Increase the temperature to encourage more creative and original outputs.
D.Fine-tune the model on the company's own original scripts to bias it toward unique content.
AnswerA

Filtered decoding, such as Google Cloud's safety filters or custom blocklists, can prevent the model from outputting specific phrases or passages that match known copyrighted material. By integrating a filter that checks generated text against a database of protected content, the company can reduce the risk of unintentional infringement. This is a targeted mitigation for the described problem.

Why this answer

Filtered decoding or output filtering is a proactive measure that scans generated text for matches with known copyrighted material and blocks or alters those outputs. This directly addresses the risk of the model reproducing protected content. Other techniques like temperature adjustment or repetition penalties do not specifically target copyright infringement and may not be effective.

Exam trap

The trap here is thinking that increasing randomness or fine-tuning inherently prevents copyright issues, when a targeted filter is needed to block known protected content.

← PreviousPage 2 of 3 · 156 questions totalNext →

Ready to test yourself?

Try a timed practice session using only Fundamentals of Generative AI questions.