AIF-C01 Applications of Foundation Models — 100 Questions

Question 1

A healthcare company is using Amazon Bedrock to summarize patient notes. The compliance team requires that no patient data is used to improve the underlying foundation model. Which configuration should the team choose?

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

Accepted Answer

Disable model training data logging in the AWS console.. Disabling model training data logging in the AWS console prevents Amazon Bedrock from using customer inference data to improve the underlying foundation model. This setting ensures compliance with the requirement that no patient data is used for model training, as Bedrock offers a specific toggle to opt out of data sharing for model improvement.

Answer

Enable data encryption in transit and at rest.

Answer

Use a different foundation model from a different provider.

Answer

Configure a VPC endpoint for Amazon Bedrock.

Question 2

Refer to the exhibit. A user invokes Claude v2 using the AWS CLI. The response is truncated. What is the most likely cause?

Accepted Answer

The prompt includes a stop sequence 'Assistant:'.. The prompt includes the stop sequence 'Assistant:', which causes the model to halt generation as soon as it encounters that token sequence. In Claude v2, stop sequences are used to control the output length and structure; when the model generates the exact stop sequence, it truncates the response at that point, even if more content could have been produced.

Answer

The AWS CLI is missing the --endpoint-url parameter.

Answer

The max_tokens_to_sample is too low.

Answer

The model does not support this use case.

Question 3

A financial services company is using Amazon Bedrock to generate investment summaries. They want to ensure that the model outputs are factually accurate and based on the latest market data. Which combination of services should they use to achieve this? (Select TWO)

Accepted Answer

Amazon Aurora with the pgvector extension. Amazon Aurora with the pgvector extension (Option D) enables storing and querying vector embeddings directly within a PostgreSQL-compatible database, which is essential for Retrieval-Augmented Generation (RAG). When combined with Amazon Bedrock Knowledge Bases (Option E), it allows the company to retrieve the most current market data as vector embeddings, ensuring the generated investment summaries are grounded in factual, up-to-date information rather than relying solely on the model's static training data.

Answer

Amazon SageMaker Ground Truth for data labeling

Answer

Amazon DynamoDB as the knowledge base store

Answer

Amazon Kendra for indexing the knowledge base

Question 4

A healthcare organization is using Amazon Bedrock to analyze medical images and generate radiology reports. They need to comply with HIPAA regulations and ensure patient data is not used for model training. Which configuration should they use?

Accepted Answer

Use a provisioned throughput model with data isolation. Provisioned Throughput with data isolation in Amazon Bedrock ensures that the customer's inference data (including patient medical images and reports) is not used for any model training or service improvement, and it provides a dedicated, isolated environment that meets HIPAA compliance requirements. This configuration guarantees that patient data remains within the customer's AWS account and is not shared with other customers or used to improve the base model.

Answer

Fine-tune the model using a custom dataset and deploy as a custom model

Answer

Use the on-demand model through Amazon Bedrock

Answer

Use a third-party model hosted outside of AWS

Question 5

A developer is using Amazon Bedrock to generate code snippets. The model often produces insecure code. Which prompt engineering technique is MOST effective to improve security?

Accepted Answer

Include specific instructions to avoid common security vulnerabilities. Directly instructing the model to avoid specific security vulnerabilities (e.g., SQL injection, buffer overflows) is the most explicit and effective way to constrain the output. Amazon Bedrock models respond well to clear, imperative instructions in the system prompt or user message, making this a direct application of prompt engineering for safety. Chain-of-thought or few-shot examples may improve reasoning or style but do not guarantee the model will avoid insecure patterns unless explicitly told to do so.

Answer

Use chain-of-thought prompting to step through the code

Answer

Provide few-shot examples of secure code

Answer

Set max_tokens to a low value to limit output

Question 6

A company is using Amazon Bedrock to build a text-to-SQL application. They want to ensure that the generated SQL queries are valid and safe. Which approach is BEST?

Accepted Answer

Configure a guardrail to filter and validate the generated SQL. Amazon Bedrock guardrails provide a native, configurable mechanism to filter and validate model outputs, including SQL queries, against defined policies such as regex patterns, denied topics, and content filters. This approach directly addresses both validity and safety without requiring additional model training or external validation services, making it the most integrated and efficient solution for ensuring generated SQL is syntactically correct and free of harmful operations like DROP or DELETE.

Answer

Fine-tune the model on a dataset of valid SQL queries

Answer

Use a separate model to validate the SQL after generation

Answer

Limit the max_tokens to 50 to reduce complexity

Question 7

A company is using Amazon Bedrock to generate product descriptions. They notice that the model sometimes produces descriptions that contain factual errors about the products. Which TWO actions should they take to improve factual accuracy?

Accepted Answer

Implement Retrieval Augmented Generation (RAG) with a product knowledge base. Retrieval Augmented Generation (RAG) grounds the model's output in a curated product knowledge base, allowing it to retrieve and cite authoritative facts during generation. This directly reduces hallucinations by ensuring the model references verified data rather than relying solely on its parametric memory.

Answer

Reduce the temperature parameter to 0.1

Answer

Increase the max_tokens to allow longer descriptions

Answer

Use human reviewers to correct errors after generation

Question 8

A company wants to classify customer emails into categories (e.g., complaint, inquiry, feedback) using a foundation model. Which approach is MOST efficient?

Accepted Answer

Use Amazon Comprehend for custom classification. Amazon Comprehend provides a managed custom classification API that is purpose-built for text classification tasks like categorizing emails. It requires only a small set of labeled data to train a custom classifier, eliminating the need to manage infrastructure or fine-tune large models, making it the most efficient choice for this specific use case.

Answer

Train a custom model using Amazon SageMaker

Answer

Fine-tune a large language model on labeled emails

Answer

Use Amazon Lex with a classifier intent

Question 9

A company is using Amazon Bedrock to generate marketing copy. They want to evaluate the quality of the generated text. Which metric is MOST suitable for assessing the relevance and coherence of the content?

Accepted Answer

ROUGE-N. ROUGE-N (Recall-Oriented Understudy for Gisting Evaluation) measures the overlap of n-grams between generated text and reference text, making it suitable for assessing relevance and coherence in content generation tasks like marketing copy. It evaluates how well the generated text captures key phrases and maintains logical flow, which aligns with the need to assess content quality beyond simple factual accuracy.

Answer

Accuracy

Answer

Perplexity

Answer

BLEU score

Question 10

A company is developing a chatbot using Amazon Bedrock and wants to ensure the model's responses do not include toxic or biased language. The company has a labeled dataset of undesirable responses. Which approach should be used to fine-tune the foundation model to reduce harmful outputs?

Accepted Answer

Use reinforcement learning from human feedback (RLHF) with a reward model trained on human preferences.. Reinforcement learning from human feedback (RLHF) is the correct approach because it directly optimizes the model to avoid toxic or biased outputs by training a reward model on human-labeled preferences. The reward model scores the model's responses, and the foundation model is fine-tuned via reinforcement learning to maximize these scores, effectively reducing harmful language. This method is specifically designed to align model behavior with nuanced human values, such as avoiding toxicity, which supervised fine-tuning alone cannot guarantee.

Answer

Perform supervised fine-tuning on a curated dataset of safe responses.

Answer

Use prompt engineering to instruct the model to avoid toxic language.

Answer

Implement adversarial validation by testing against toxic inputs.

Question 11

A developer is calling the Amazon Bedrock InvokeModel API to generate text with the AI21 Labs Jurassic-2 Mid model. The API call includes a maxTokens parameter, but the request fails with the error shown in the exhibit. What is the most likely cause of this error?

Accepted Answer

The parameter name is incorrect; the model expects 'maxTokens' with a capital T.. The AI21 Labs Jurassic-2 Mid model expects the parameter name 'maxTokens' with a capital 'T' (camelCase). The error occurs because the API request used a different casing (e.g., 'maxtokens' or 'max_tokens'), which the model's schema does not recognize. Amazon Bedrock's InvokeModel API passes parameters directly to the model, so parameter names must match the model's exact specification.

Answer

The API request is missing a required parameter such as 'prompt'.

Answer

The AWS region does not support the AI21 Labs model.

Answer

The value of 'maxTokens' exceeds the model's maximum limit.

Question 12

An e-commerce company uses a foundation model to generate personalized email subject lines. The marketing team notices that the subject lines sometimes contain product recommendations that are out of stock. Which action would best reduce the generation of out-of-stock recommendations without retraining the model?

Accepted Answer

Add a system prompt that explicitly instructs the model to only recommend products that are in stock.. Adding a system prompt that explicitly instructs the model to only recommend in-stock products directly constrains the model's output at inference time without requiring retraining. This leverages the model's instruction-following capability to filter its generated content based on the provided context, which is a lightweight and immediate solution.

Answer

Implement a post-processing step to replace out-of-stock recommendations with in-stock alternatives.

Answer

Fine-tune the model on a dataset of past successful subject lines that only include in-stock products.

Answer

Use a retrieval-augmented generation (RAG) approach to retrieve a list of in-stock products and include it in the prompt.

Question 13

A company wants to use a foundation model to classify customer feedback into positive, neutral, negative. They have a small labeled dataset. What approach yields best results?

Accepted Answer

Fine-tune a foundation model on their dataset. Fine-tuning a foundation model on a small labeled dataset allows the model to adapt its pre-trained knowledge specifically to the company's sentiment classification task, achieving higher accuracy than zero-shot or generic API approaches. Fine-tuning adjusts the model's weights using the labeled examples, making it sensitive to domain-specific language and nuance in customer feedback, which is critical for a three-class sentiment task.

Answer

Use a pre-built sentiment analysis API

Answer

Fine-tune a foundation model on their dataset

Answer

Use zero-shot classification

Question 14

A developer wants to quickly experiment with multiple foundation models using a single API. Which service provides this capability?

Accepted Answer

Amazon Bedrock. Amazon Bedrock is a fully managed service that provides a single API to access and experiment with multiple foundation models from leading AI providers like AI21 Labs, Anthropic, Cohere, Meta, Stability AI, and Amazon itself. This allows developers to quickly test different models without managing underlying infrastructure or learning separate APIs for each provider.

Answer

AWS Lambda

Answer

Amazon Bedrock

Answer

Amazon SageMaker Studio

Question 15

A data scientist is fine-tuning a foundation model on a custom dataset using Amazon SageMaker. After training, the model shows high accuracy on training data but poor on validation. Which action should be taken?

Accepted Answer

Reduce training epochs or add regularization. The model is overfitting, as indicated by high training accuracy but poor validation performance. Reducing training epochs or adding regularization (e.g., L1/L2 weight decay) directly addresses overfitting by limiting the model's capacity to memorize noise. In Amazon SageMaker, this can be implemented via hyperparameter tuning or by modifying the training script to include regularization terms.

Answer

Add dropout layers

Answer

Increase learning rate

Answer

Use a different foundation model

Question 16

Which THREE of the following are factors to consider when selecting a foundation model for a text generation task?

Accepted Answer

Pricing per token. Option B (Pricing per token) is correct because cost scales directly with token usage in text generation, so per-token pricing is a key economic factor when choosing a foundation model. Option C (Model size in parameters) is correct because parameter count affects capability, latency, and infrastructure requirements, all of which matter for a text generation workload. Option D (Training data source and diversity) is correct because the provenance and breadth of training data influence output quality, bias, and domain suitability for the generation task. Option A (Supported output modalities) is not a primary factor here since the task is specifically text generation, making non-text modalities irrelevant. Option E (Availability of automatic scaling) is not a foundation-model selection factor but rather a deployment/platform concern handled by the serving infrastructure.

Answer

Supported output modalities

Answer

Availability of automatic scaling

Question 17

Which FOUR of the following are benefits of using Amazon Bedrock for foundation models?

Accepted Answer

Ability to fine-tune models. Amazon Bedrock is a fully managed, serverless service, so option E is correct because you consume foundation models without provisioning or managing any infrastructure, paying only for what you use. Option B is correct because Bedrock exposes a single, unified API (via the AWS SDK, CLI, or console) that lets you invoke models from multiple providers such as Amazon Titan, Anthropic Claude, AI21 Labs, Cohere, Meta Llama, Mistral, and Stability AI. Option A is correct because Bedrock supports customizing foundation models through fine-tuning (for supported models) as well as continued pre-training and Retrieval Augmented Generation, so you can adapt a model to your domain data. Option D is correct because Bedrock integrates with Amazon CloudWatch for metrics/logs and AWS CloudTrail for API activity auditing, and it supports guardrails, model invocation logging, and IAM-based access control for governance. Option C is not correct because no generative AI service can guarantee output accuracy; foundation models can hallucinate, and Bedrock provides tools like Guardrails and RAG to improve reliability but does not promise correct outputs.

Answer

Guaranteed output accuracy

Question 18

Refer to the exhibit. A developer deploys this CloudFormation stack but the agent fails to query the knowledge base. What is a likely cause?

Accepted Answer

The agent role does not have permissions to invoke the knowledge base. The agent role must have an IAM policy that grants the `bedrock:Retrieve` and `bedrock:RetrieveAndGenerate` permissions on the knowledge base. Without these permissions, the agent cannot invoke the knowledge base, even if the KnowledgeBaseId is correctly passed and the embedding model is available.

Answer

The KnowledgeBaseId is not passed correctly

Answer

The embedding model is not available in the region

Answer

The OpenSearch collection type should be SEARCH not VECTORSEARCH

Question 19

A company uses Amazon SageMaker JumpStart to deploy a foundation model. They want to fine-tune the model on their own dataset. Which SageMaker capability should they use?

Accepted Answer

SageMaker Studio Classic. SageMaker Studio Classic provides a fully integrated development environment (IDE) for machine learning, including built-in support for fine-tuning foundation models deployed via JumpStart. It allows you to access JumpStart model hubs, configure training jobs with custom datasets, and manage the fine-tuning process directly from the Studio interface, making it the correct choice for this task.

Answer

SageMaker Managed Spot Training

Answer

SageMaker Canvas

Answer

SageMaker Autopilot

Question 20

A company is using Amazon Bedrock to generate images. They want to ensure the outputs comply with content policies. Which TWO AWS services can help? (Choose two.)

Accepted Answer

Amazon Augmented AI (A2I). Amazon Augmented AI (A2I) is correct because it provides a human review workflow that can route Bedrock-generated images (or their metadata) to human reviewers who validate compliance with content policies before the output is released. Amazon Rekognition is correct because its DetectModerationLabels API automatically identifies unsafe or policy-violating content categories (e.g., explicit nudity, violence, drugs) in images, enabling automated filtering of Bedrock outputs. Amazon GuardDuty is not relevant because it is a threat-detection service for AWS accounts, workloads, and logs (e.g., CloudTrail, VPC Flow Logs), not image content moderation. AWS WAF does not belong because it filters HTTP/S web traffic at the application layer (e.g., SQL injection, XSS) and cannot inspect image content. Amazon Comprehend is incorrect because it is an NLP service for text (sentiment, entities, moderation of text), not image content analysis.

Answer

Amazon GuardDuty

Answer

AWS WAF

Answer

Amazon Comprehend