AIF-C01 Practice Test 30 — 15 Questions

Question 1

A company is fine-tuning an Amazon Titan Text model on custom data using Amazon Bedrock. They want to ensure the fine-tuned model retains general language capabilities while learning domain-specific knowledge. Which THREE best practices should they follow? (Select THREE.)

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

Accepted Answer

Monitor validation loss to detect overfitting and stop training early if needed. Option C is correct because monitoring validation loss during fine-tuning lets you detect when the model begins to overfit the domain data and apply early stopping, preserving generalization to broader language tasks. Option D is correct because using a low learning rate makes smaller weight updates, which reduces the risk of catastrophic forgetting of the model's pre-trained general language capabilities while it absorbs domain knowledge. Option E is correct because mixing general and domain-specific examples in the training dataset helps the fine-tuned model retain broad language understanding alongside the new specialized knowledge. Option A is not appropriate here because freezing the first few layers is a parameter-efficient technique aimed at reducing overfitting and compute, not a stated best practice for preserving general capabilities in Amazon Bedrock Titan fine-tuning. Option B is incorrect because training only on domain-specific data maximizes specialization at the expense of general language ability, which is exactly the outcome the company wants to avoid.

Answer

Freeze the first few layers of the model to prevent overfitting

Answer

Train only on domain-specific data to maximize accuracy

Question 2

A company needs to store large amounts of unstructured training data (images, videos) in a cost-effective manner while ensuring low-latency retrieval for training jobs running on Amazon SageMaker. Which storage solution should be used?

Accepted Answer

Amazon S3. Amazon S3 is the correct choice because it is designed for cost-effective, scalable storage of unstructured data (images, videos) and integrates natively with Amazon SageMaker for low-latency data retrieval during training jobs. S3 provides high throughput and can be accessed directly from SageMaker training instances without the need for file system mounting, making it ideal for large-scale ML workloads.

Answer

Amazon EFS

Answer

Amazon RDS

Answer

Amazon EBS

Question 3

A company is deploying an LLM for generating marketing copy. They want to reduce the risk of hallucinations and ensure the content is factually accurate. Which TWO approaches should they implement?

Accepted Answer

Use Retrieval-Augmented Generation (RAG) with a verified knowledge base. Option A, Retrieval-Augmented Generation (RAG) with a verified knowledge base, is correct because RAG grounds the model's output in retrieved, authoritative documents rather than relying solely on parametric memory, which directly reduces hallucination risk and improves factual accuracy for marketing copy. Option B, implementing human review of generated content before publication, is correct because a human-in-the-loop check catches residual factual errors, unsupported claims, and brand/compliance issues that automated grounding may miss before content goes live. Option C is not correct because model size alone does not determine hallucination rate; smaller models can hallucinate as much or more, and the scenario calls for grounding and verification, not a smaller model. Option D is not correct because disabling guardrails removes safety and accuracy controls, increasing rather than reducing the risk of harmful or false output. Option E is not correct because increasing the temperature parameter raises randomness and creativity, which typically increases hallucination and factual drift instead of improving accuracy.

Answer

Use a smaller LLM to reduce the chance of hallucinations

Answer

Disable all guardrails to avoid interference

Answer

Increase the temperature parameter to generate more diverse outputs

Question 4

Which TWO actions should a data scientist take to evaluate fairness of a binary classification model using Amazon SageMaker Clarify? (Choose two.)

Accepted Answer

Use post-training bias metrics like Difference in Positive Proportions. Option A is correct because SageMaker Clarify's post-training bias metrics, such as Difference in Positive Proportions (DPP), compare predicted outcomes across facets (e.g., gender or age groups) to quantify disparate impact in the model's actual predictions, which is essential for evaluating fairness of a binary classifier. Option D is correct because pre-training bias metrics like Class Imbalance (CI) measure skew in the underlying training data's label distribution across facets before any model is trained, helping detect whether the data itself could lead to biased outcomes. Together, these two metric types cover both data-level and model-level fairness assessment as Clarify is designed to do. Option B is not a Clarify fairness evaluation action but a data preprocessing technique, and balancing data does not by itself measure bias. Option C, SHAP values, explains feature importance/attribution for interpretability, not fairness metrics. Option E, data quality monitoring on unlabeled data, addresses data drift/quality issues and does not evaluate model fairness.

Answer

Ensure the training dataset is balanced by resampling

Answer

Generate SHAP values for feature importance

Answer

Run a data quality monitoring job on unlabeled data

Question 5

Which THREE considerations are important when implementing responsible AI for a production NLP system? (Choose three.)

Accepted Answer

Continuously monitor model outputs for bias and drift. Option B is correct because responsible AI in production requires ongoing monitoring of model outputs to detect bias and data/concept drift, since model behavior can degrade or become unfair as real-world data changes over time. Option D is correct because model cards are a standard transparency artifact that document a model's intended use, performance metrics across relevant subgroups, and known limitations, enabling informed and accountable use by stakeholders. Option E is correct because embedding bias detection as an automated gate in the CI/CD pipeline ensures every model update is evaluated for fairness regressions before deployment, making responsible AI a repeatable part of the ML lifecycle rather than an afterthought. Option A does not belong because FDA approval applies to regulated medical devices and clinical AI, not to general production NLP systems. Option C does not belong because encrypting training code at rest is a generic security control and is not a responsible-AI consideration such as fairness, transparency, or accountability.

Answer

Obtain FDA approval for the model

Answer

Apply encryption at rest for all training code

Question 6

Which of the following is a key principle of inclusive design in AI?

Accepted Answer

Ensure the AI system is usable by people with diverse abilities and backgrounds. Inclusive design in AI means creating systems that are accessible and usable by people with diverse abilities, backgrounds, and characteristics. This involves considering a wide range of users during design, including those with disabilities, different languages, and varied cultural contexts. Option C captures this principle by emphasizing usability across diverse abilities and backgrounds.

Answer

Deploy AI systems without user testing

Answer

Use only data from a single demographic group

Answer

Maximize model complexity for better performance

Question 7

A company uses a text generation model to produce legal documents. They want to minimize the environmental impact of training and inference. Which THREE approaches should they consider?

Accepted Answer

Train the model on a smaller but representative dataset. Option A is correct because training on a smaller but representative dataset reduces the total compute cycles, energy consumption, and carbon emissions required for model training while still maintaining acceptable model quality. Option C is correct because distilled or otherwise more efficient architectures require fewer parameters and FLOPs per inference, directly lowering the energy and hardware resources needed for both training and serving. Option E is correct because SageMaker managed spot training uses spare AWS capacity and automatically checkpoints/resumes jobs, which improves utilization and reduces idle compute, thereby lowering the environmental footprint. Option B is not appropriate because replicating training data across multiple AWS Regions increases storage, replication traffic, and energy use without reducing the model's environmental impact. Option D is not appropriate because on-demand instances do not inherently reduce energy consumption and can leave resources idle; they address performance consistency, not environmental efficiency.

Answer

Store all training data in multiple AWS Regions for redundancy

Answer

Perform all training on on-demand instances to ensure consistent performance

Question 8

A company needs to fine-tune a foundation model on a large dataset of proprietary documents. They are concerned about data privacy and want to ensure that no data leaves their AWS account. Which Amazon Bedrock feature should they use?

Accepted Answer

Custom Model Import. Custom Model Import allows you to import a pre-trained foundation model (e.g., from Hugging Face) into your own AWS environment and fine-tune it using your proprietary data entirely within your account. This ensures that no data leaves your AWS account, addressing the privacy requirement directly.

Answer

Guardrails for Amazon Bedrock

Answer

Provisioned Throughput

Answer

Knowledge Bases for Amazon Bedrock

Question 9

A company needs to secure data used for training machine learning models in Amazon SageMaker. Which TWO actions should be taken to protect data at rest and in transit? (Choose TWO.)

Accepted Answer

Enable default encryption on S3 buckets using AWS KMS. Enabling default encryption on S3 buckets using AWS KMS ensures that all objects stored in the bucket are encrypted at rest. This is a fundamental security practice for protecting training data in SageMaker, as SageMaker reads data from S3. Using AWS KMS allows you to manage and audit encryption keys, providing fine-grained control over data protection.

Answer

Use AWS Certificate Manager (ACM) to generate certificates for all SageMaker endpoints

Answer

Disable root access on SageMaker notebook instances

Answer

Use SageMaker Model Registry to track model versions

Question 10

A company is using Amazon SageMaker to train machine learning models. The security team wants to ensure that the training data is encrypted at rest and that the SageMaker notebook instances cannot access the internet. Which TWO actions should the company take? (Choose TWO.)

Accepted Answer

Enable S3 server-side encryption with AWS KMS (SSE-KMS) for the training data bucket. Enabling S3 server-side encryption with AWS KMS (SSE-KMS) ensures that the training data stored in the S3 bucket is encrypted at rest. This satisfies the security team's requirement for data encryption at rest, as SSE-KMS provides envelope encryption with a customer-managed or AWS-managed KMS key, giving the company control over the encryption keys and auditability via AWS CloudTrail.

Answer

Create an AWS CloudTrail trail to log all S3 data events

Answer

Enable encryption at rest for the SageMaker endpoint using the AWS Management Console

Answer

Use AWS Security Token Service (STS) to generate temporary credentials for the notebook instance

Question 11

A healthcare AI system predicts patient diagnoses. The data collection process primarily samples from urban hospitals, leading to underrepresentation of rural populations. Which type of bias is this, and what is the most effective mitigation strategy?

Accepted Answer

Representation bias; collect additional data from rural hospitals. This is representation bias, which occurs when the training data underrepresents certain subgroups (rural populations) relative to the population the model will serve. The most effective mitigation is to collect additional data from the underrepresented group (rural hospitals) so the model learns their patterns and does not systematically misdiagnose them.

Answer

Historical bias; reweight urban samples to reduce their influence

Answer

Aggregation bias; use regularization to simplify the model

Answer

Measurement bias; apply data augmentation to rural records

Question 12

A company uses an LLM to generate medical advice. They are concerned about hallucinations and want to implement safeguards. Which TWO techniques should they prioritize?

Accepted Answer

Implement Retrieval-Augmented Generation (RAG) using a verified medical knowledge base. Option C is correct because Retrieval-Augmented Generation (RAG) grounds the model's responses in a verified medical knowledge base, so answers are conditioned on authoritative, up-to-date source documents rather than the model's parametric memory, which directly reduces fabricated or unsupported claims. Option D is correct because a human-in-the-loop workflow with Amazon Augmented AI (A2I) routes high-risk or low-confidence responses to qualified reviewers before they reach patients, providing a safety net that catches residual hallucinations that automated grounding cannot eliminate. The other options do not belong: lowering temperature to 0 (A) only makes sampling more deterministic and does not prevent the model from confidently stating false facts; fine-tuning on a general medical text dataset (B) can improve domain fluency but does not guarantee factual accuracy or verifiability and may even reinforce errors; and Bedrock Guardrails content filters (E) are designed to block harmful, toxic, or policy-violating content, not to verify the factual correctness of medical advice.

Answer

Reduce the temperature parameter to 0

Answer

Fine-tune the LLM on a general dataset of medical texts

Answer

Use Bedrock Guardrails with a content filter for medical topics

Question 13

Which AWS service allows teams to collaborate on building generative AI applications with a visual interface, share prompts, and manage prompt versions?

Accepted Answer

Amazon Bedrock Studio. Amazon Bedrock Studio is the correct service because it provides a visual interface specifically designed for teams to collaborate on building generative AI applications. It allows users to share prompts, manage prompt versions, and iterate on foundation model configurations in a team-based environment, directly matching the question's requirements.

Answer

Amazon Bedrock Playground

Answer

AWS CodeCommit

Answer

Amazon SageMaker Studio

Question 14

A team is using Amazon SageMaker to train a deep learning model. They notice that the training loss decreases steadily but the validation loss starts increasing after 10 epochs. Which technique should they apply to address this issue?

Accepted Answer

Apply early stopping. The scenario describes overfitting, where the model memorizes training data but fails to generalize to validation data. Early stopping halts training when validation loss stops improving, preventing overfitting while preserving the best model weights. This is a standard regularization technique in SageMaker training jobs, configurable via the `use_early_stopping` parameter in the `Estimator` or `HyperparameterTuner`.

Answer

Reduce the batch size

Answer

Increase the learning rate

Answer

Add more layers to the network

Question 15

An organization uses Amazon Macie to discover sensitive data in S3. The compliance team needs to receive notifications when Macie finds PII. Which AWS service should they use to send notifications?

Accepted Answer

Amazon Simple Notification Service (SNS). Amazon Macie publishes findings to Amazon EventBridge, which can trigger a target such as an SNS topic. SNS then pushes notifications (e.g., email, SMS, HTTP) to subscribers. This decoupled architecture allows the compliance team to receive real-time alerts when PII is detected, without needing to poll or manage compute resources.

Answer

AWS Lambda

Answer

Amazon CloudWatch

Answer

Amazon Simple Email Service (SES)