AIF-C01 Practice Test 38 — 15 Questions

Question 1

An e-commerce company uses an LLM to generate product descriptions. They observe that occasionally the model outputs factually incorrect information about products. What is the term for this phenomenon?

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

Accepted Answer

Hallucination. Hallucination is the term for an LLM generating fluent, confident output that is factually incorrect or unsupported by its training data — exactly what happens when the model invents product specifications, prices, or features. It occurs because the model predicts statistically plausible tokens rather than retrieving verified facts. This is a well-known limitation of generative AI and the reason techniques like RAG and grounding are used to reduce it.

Answer

Bias amplification

Answer

Overfitting

Answer

Concept drift

Question 2

A machine learning engineer is implementing a RAG system using Amazon Bedrock and a vector database. They need to chunk a large set of PDF documents before embedding. Which THREE considerations are important for chunking strategy? (Select THREE.)

Accepted Answer

Overlap between consecutive chunks. Options A, B, and C are the three core design parameters of any chunking strategy. A is correct because adding overlap between consecutive chunks preserves context that would otherwise be lost at chunk boundaries, improving retrieval quality when a relevant passage spans two chunks. B is correct because the splitting method (sentence, paragraph, semantic, or fixed token count) determines how coherently meaning is preserved within each chunk and directly affects embedding quality. C is correct because chunk size in tokens controls the trade-off between retrieval granularity and the amount of context each embedding captures; too small loses context, too large dilutes relevance. Option D is not a chunking-strategy consideration per se — the embedding model's maximum input length is a hard constraint that caps chunk size, but it is a model property rather than a strategic choice. Option E is about document metadata and filtering, which relates to retrieval and indexing, not to how documents are split into chunks.

Answer

The embedding model's maximum input length

Answer

The metadata associated with each document

Question 3

A healthcare startup is deploying an AI model to assist with diagnosis. They want to comply with the EU AI Act, which classifies medical AI as high-risk. Which THREE requirements must they fulfill? (Choose three.)

Accepted Answer

Provide technical documentation and logs for traceability. Under the EU AI Act, high-risk AI systems such as medical diagnostic AI must satisfy a set of mandatory requirements, and option A is correct because providers must supply technical documentation and maintain automatically generated logs that enable traceability and post-market monitoring of the system's operation. Option B is correct because Article 9 requires a risk management system that is established, implemented, documented, and maintained as a continuous iterative process throughout the entire lifecycle of the high-risk AI system. Option C is correct because high-risk systems must be designed to allow effective human oversight, enabling humans to prevent or minimize risks to health, safety, or fundamental rights, which is especially critical in a clinical diagnosis context. Option D is not a stated EU AI Act requirement; the Act addresses data governance, quality, and representativeness rather than mandating minimization of training data to only necessary data. Option E is incorrect because allowing the model to make final decisions without human review directly contradicts the human oversight requirement for high-risk AI systems.

Answer

Minimize the amount of training data to only necessary data

Answer

Allow the model to make final decisions without human review

Question 4

A retail company uses a recommendation system that occasionally suggests inappropriate products to minors. Which responsible AI practice should be applied?

Accepted Answer

Implement human review of flagged recommendations. The correct practice is to implement human review of flagged recommendations. This aligns with the responsible AI principle of accountability, where automated systems must have oversight mechanisms to catch and correct inappropriate outputs, especially when minors are involved. Human-in-the-loop (HITL) validation ensures that edge cases or subtle context (e.g., age-inappropriate product suggestions) are caught before they reach end users, rather than relying solely on automated filters or feedback loops.

Answer

Rely solely on user feedback to improve

Answer

Disable the recommendation system entirely

Answer

Increase the volume of training data

Question 5

A healthcare company is deploying a machine learning model on Amazon SageMaker to analyze patient records. The model requires access to a DynamoDB table containing patient data. Which combination of AWS services and features should the company use to restrict access to only the necessary resources?

Accepted Answer

Create an IAM role with a policy granting read-only access to the specific DynamoDB table and attach it to the SageMaker notebook instance. It follows the AWS principle of least privilege by creating an IAM role with a policy that grants read-only access to the specific DynamoDB table, then attaching that role to the SageMaker notebook instance. This ensures the notebook can only perform read operations on the required table without exposing long-term credentials or granting broader permissions.

Answer

Attach a DynamoDB resource-based policy to the table allowing access from the SageMaker notebook

Answer

Store AWS access keys in the notebook and use those credentials to access DynamoDB

Answer

Launch the SageMaker notebook in a VPC with a security group that allows access to DynamoDB

Question 6

A legal firm uses Amazon Bedrock to generate contract summaries. They want to evaluate the quality of summaries against human-written reference summaries. The evaluation should capture both the overlap of n-grams and the semantic similarity. Which combination of automated metrics is MOST appropriate?

Accepted Answer

ROUGE and BERTScore. ROUGE measures n-gram overlap (recall-oriented) for summarization, while BERTScore uses contextual embeddings to capture semantic similarity. BLEU is for translation. Combining ROUGE and BERTScore gives both lexical and semantic evaluation.

Answer

Exact match and F1 score

Answer

ROUGE and BLEU

Answer

BLEU and BERTScore

Question 7

A financial services firm needs to ensure that all calls to Amazon Bedrock APIs are logged for audit purposes. Which AWS service should they enable to capture API calls?

Accepted Answer

AWS CloudTrail. AWS CloudTrail records API activity in AWS accounts, including Bedrock API calls, providing audit logs.

Answer

Amazon S3 server access logs

Answer

Amazon CloudWatch Logs

Answer

AWS Config

Question 8

A startup company is developing an e-commerce platform and wants to use Amazon Bedrock to generate product descriptions automatically. They have a small team of developers who are not machine learning experts. The product catalog is stored in a DynamoDB table, and each product has attributes like name, category, price, and a brief description. The company wants the generated descriptions to reflect the unique brand voice, which is documented in a few internal style guides stored as PDF files in Amazon S3. They need a solution that allows them to quickly test the approach without significant infrastructure changes or model training. The development team is familiar with AWS SDKs and want to minimize ongoing maintenance. The team has already set up a Bedrock foundation model (Claude) and can make API calls. They tested simple prompts but the output lacked the brand's informal yet professional tone. They want to incorporate examples from the style guides directly into the prompt without retraining. The team fears that including the entire style guide in each prompt would exceed token limits and increase costs. Which approach should they take to effectively incorporate the brand voice with minimal changes?

Accepted Answer

Use Amazon Bedrock with a custom prompt template that includes a few representative examples from the style guides as few-shot examples in the system prompt.. Amazon Bedrock supports few-shot prompting, where you include a few representative examples in the prompt to guide the model's style without retraining. By extracting a few examples from the style guides and incorporating them into the system prompt, the model can learn the desired tone and apply it to new product descriptions. This approach requires minimal changes, no model training, and keeps token usage manageable by not including the entire style guide.

Answer

Fine-tune the foundation model using the style guides with Amazon Bedrock Custom Models.

Answer

Concatenate all style guide PDFs into a single text and include it in every prompt.

Answer

Use Amazon Comprehend to analyze the style guides and extract a list of keywords to include in the prompt.

Question 9

During model training, a data scientist notices that the model performs very well on the training data but poorly on the test data. The scientist suspects high variance. Which technique is MOST likely to reduce the variance and improve test performance?

Accepted Answer

Apply regularization (e.g., L1 or L2). High variance indicates the model is overfitting to the training data, capturing noise rather than the underlying pattern. Regularization (L1/L2) adds a penalty to the loss function for large coefficients, effectively constraining the model complexity and reducing variance, which improves generalization to unseen test data.

Answer

Lower the learning rate

Answer

Increase the number of features

Answer

Decrease the amount of training data

Question 10

A financial services company is deploying a machine learning model to approve loan applications. To comply with regulatory requirements, they must ensure the model does not discriminate based on race. They have historical data that may contain bias. Which AWS service can help detect and measure bias in the dataset and model predictions?

Accepted Answer

Amazon SageMaker Clarify. Amazon SageMaker Clarify is specifically designed to detect and measure bias in datasets and model predictions. It provides tools to compute bias metrics such as disparate impact (DI) and difference in conditional acceptance (DCA) before training and after deployment. This directly addresses the regulatory requirement to ensure the model does not discriminate based on race. Clarify integrates with SageMaker to provide bias reports and explanations, making it the correct choice for bias detection and measurement.

Answer

AWS CloudTrail

Answer

Amazon SageMaker Model Monitor

Answer

AWS Identity and Access Management (IAM)

Question 11

A data scientist wants to fine-tune a foundation model on a specific domain dataset using Amazon SageMaker. Which built-in SageMaker feature can simplify the training process?

Accepted Answer

SageMaker JumpStart. SageMaker JumpStart provides pre-trained foundation models and built-in training scripts that simplify fine-tuning on custom datasets. It handles the underlying infrastructure, hyperparameter configurations, and model deployment, allowing the data scientist to focus on the domain-specific data rather than writing custom training loops or managing SageMaker training jobs manually.

Answer

SageMaker Neo

Answer

SageMaker Canvas

Answer

SageMaker Ground Truth

Question 12

A developer is using Amazon Bedrock to create a chatbot. They want to ensure the bot does not generate toxic or offensive content. Which feature should they enable?

Accepted Answer

Enable content filtering on the Bedrock model.. Amazon Bedrock provides built-in content filtering capabilities that can be enabled at the model invocation level to automatically detect and block toxic or offensive content in both input prompts and generated responses. This feature uses predefined safety filters (e.g., hate, insults, sexual content, violence) and is the most direct and managed way to prevent harmful outputs without requiring custom development.

Answer

Use careful prompt engineering to avoid toxic responses.

Answer

Fine-tune the model on a dataset of safe responses.

Answer

Implement external response validation using a third-party API.

Question 13

A data scientist wants to host a pre-trained model on Amazon SageMaker for real-time inference with minimal latency. Which approach should they use?

Accepted Answer

Deploy the model on a SageMaker real-time endpoint. SageMaker real-time endpoints are designed for low-latency, synchronous inference. They keep the model loaded and ready to respond to individual requests, making them ideal for real-time applications where minimal latency is critical.

Answer

Run inference using AWS Lambda with the model packaged as a container

Answer

Use SageMaker batch transform

Answer

Create a SageMaker asynchronous inference endpoint

Question 14

A healthcare startup is using Amazon SageMaker to train a model on patient data. They need to ensure that the training data does not contain any personally identifiable information (PII) before being used. Which AWS service can automatically detect and report PII in the data stored in S3?

Accepted Answer

Amazon Macie. Amazon Macie is a fully managed data security and privacy service that uses machine learning to automatically discover, classify, and protect sensitive data stored in Amazon S3. It can detect personally identifiable information (PII) such as names, addresses, and credit card numbers, and provides detailed reports. This directly addresses the requirement to detect and report PII in S3 before using it for training.

Answer

Amazon Rekognition

Answer

AWS Glue DataBrew

Answer

Amazon Comprehend Medical

Question 15

A data engineer is using Amazon SageMaker Data Wrangler to prepare tabular data for ML. Which THREE data transformations are natively supported? (Choose three.)

Accepted Answer

One-hot encoding for categorical features. SageMaker Data Wrangler natively supports one-hot encoding as a categorical-encoding transform, which converts categorical features into binary indicator columns suitable for ML models, so option A is correct. It also provides a text vectorization transform using TF-IDF (and other text transforms like bag-of-words and n-gram) to convert text fields into numeric feature vectors, making option C correct. Data Wrangler additionally allows custom transformations through the Custom Transform node, where users can write their own Pandas or PySpark code, so option D is correct. Options B and E are not native Data Wrangler tabular transforms: audio feature extraction and image resizing/normalization are handled by other AWS services or frameworks (for example, SageMaker Processing with librosa or image libraries), not by Data Wrangler's built-in transform list.

Answer

Audio feature extraction

Answer

Image resizing and normalization