A. Correct.
Unsupervised learning discovers patterns such as clusters in unlabeled data.

B. Incorrect.
Animal image recognition is typically supervised learning because the training images have labels.

C. Incorrect.
House-price prediction is supervised regression because it learns from examples with known prices.

D. Incorrect.
Text generation is a generative AI use case, not the standard example of unsupervised learning.

A. Incorrect.
R-squared is used for regression quality, not basic image classification correctness.

B. Correct.
Accuracy measures the proportion of correct predictions, which matches the requirement.

C. Incorrect.
RMSE measures numeric prediction error in regression problems.

D. Incorrect.
Learning rate is a training hyperparameter, not an evaluation metric.

A. Incorrect.
Agents can call tools and orchestrate tasks. They do not supervise model training to remove hallucinations.

B. Incorrect.
Preprocessing can improve data quality, but hallucinations are not usually fixed by finding one problematic record.

C. Correct.
Lower temperature makes output more deterministic and can reduce random, unsupported responses.

D. Incorrect.
No general foundation model is guaranteed to never hallucinate.

A. Correct.
Increasing temperature can make output more varied and creative.

B. Incorrect.
Batch size is related to processing records and training efficiency, not inference-time diversity.

C. Incorrect.
Learning rate affects training updates, not normal runtime response variation.

D. Incorrect.
Optimizer type affects training, not diversity during model inference.

A. Incorrect.
Model Cards are governance documentation, not marketing assets.

B. Correct.
Model Cards capture standardized model information such as intended use, metrics, risks, and limitations.

C. Incorrect.
Model Cards do not optimize compute or change model architecture.

D. Incorrect.
Model storage and archival are separate model registry or storage concerns.

A. Incorrect.
Architecture and hyperparameter changes are not available for most managed pre-trained model usage and do not directly encode brand voice.

B. Incorrect.
Adding layers means changing model architecture, which is unnecessary and impractical for a managed pre-trained model.

C. Correct.
Prompt engineering can guide tone, style, audience, and messaging requirements without retraining.

D. Incorrect.
Pre-training a new model is costly and unnecessary for brand-voice guidance.

A. Incorrect.
Data Wrangler is for preparing and transforming data, not a fully managed labeling workforce.

B. Correct.
SageMaker Ground Truth Plus provides managed data labeling, including workforce and workflow management.

C. Incorrect.
Transcribe converts speech to text and is unrelated to dataset labeling workflows.

D. Incorrect.
Macie discovers sensitive data in S3 and does not label ML training datasets.

A. Correct.
Amazon Macie is a managed service that detects sensitive data in S3 and can generate findings for alerting.

B. Incorrect.
A custom LLM redaction pipeline adds development and operational overhead.

C. Incorrect.
Regex patterns require custom development, maintenance, and ongoing tuning.

D. Incorrect.
User guidance does not automate detection and cannot ensure compliance.

A. Incorrect.
K-nearest neighbors is typically used for supervised classification or regression.

B. Correct.
K-means is an unsupervised clustering algorithm commonly used for customer segmentation.

C. Incorrect.
Decision trees are supervised models used for classification or regression.

D. Incorrect.
Support vector machines are supervised models and are not the standard clustering choice.

A. Correct.
On-Demand usage lets the company pay for what it uses without reserving capacity ahead of time.

B. Incorrect.
Model customization is a capability, not a pricing model for flexible usage.

C. Incorrect.
Provisioned Throughput reserves model capacity and is better for predictable workloads that need guaranteed throughput.

D. Incorrect.
Spot Instances are an Amazon EC2 pricing model, not an Amazon Bedrock inference pricing option.