AWS AI Practitioner Practice Test

A. ✓ Correct: GUARDRAIL_INTERVENED is the action value that indicates the guardrail intervened.

B. × Incorrect: NONE is incorrect because indicates that the guardrail did not intervene.

C. × Incorrect: BLOCKED is incorrect because describes a possible content-handling outcome, not the ApplyGuardrail action value in the response.

D. × Incorrect: FILTERED is incorrect because describes policy-style handling, not the response action returned for an intervention.

A. × Incorrect: Amazon Rekognition is incorrect because it is used for image analysis, not document extraction.

B. ✓ Correct: Amazon Textract can extract text and data from scanned documents efficiently.

C. × Incorrect: Amazon Comprehend is incorrect because it is designed to identify the language of a document and extract entities, but does not specialize in extracting structured data from documents.

D. × Incorrect: Amazon SageMaker is incorrect because it is used for building, training, and deploying machine learning models, not specifically for document extraction.

A. ✓ Correct: Privacy and security is because it involves obtaining, using, and protecting data and models appropriately.

B. × Incorrect: Explainability is incorrect as it pertains to understanding and communicating how an AI makes decisions, not managing data and models.

C. × Incorrect: Fairness is incorrect since it deals with equitable treatment of different groups of stakeholders rather than handling data and models.

D. × Incorrect: Safety is incorrect because it is because it focuses on reducing harmful system output and misuse, not obtaining or protecting data.

A. × Incorrect: Granting full administrative permissions to all service roles is incorrect because granting full administrative permissions increases security risks by allowing unnecessary access.

B. ✓ Correct: Assigning specific permissions only as needed based on task requirements is correct because it adheres to the principle of least privilege, limiting service roles to required permissions only.

C. × Incorrect: Using a single role for multiple unrelated services to simplify management is incorrect because using a single role for multiple unrelated services can lead to over-permissioning and increased risk.

D. × Incorrect: Removing all direct user access and relying solely on service roles is incorrect because as removing all direct user access without proper alternatives may hinder necessary administrative tasks.

A. ✓ Correct: Creating a knowledge base to store private data is an optional step when setting up an agent.

B. × Incorrect: Defining action groups is incorrect because as defining action groups that the agent can perform is essential for its functionality.

C. × Incorrect: Associating a knowledge base with the agent is incorrect because since associating a knowledge base with the agent is crucial for augmenting performance and accuracy.

D. × Incorrect: Customizing prompt templates is incorrect because to enhance the agent's behavior is necessary but optional.

A. × Incorrect: Model cards is incorrect because it is used to specify risk ratings but do not directly track endpoint performance.

B. × Incorrect: Workflow lineage visualization is incorrect because it is a feature, but it does not specifically track endpoint performance.

C. ✓ Correct: Endpoint performance tracking is correct because as the Model Dashboard provides comprehensive tracking of deployed models and their endpoints for real-time monitoring.

D. × Incorrect: Baseline dataset is incorrect because s are used by SageMaker Model Monitor to detect data drift, not for tracking endpoint performance.

A. × Incorrect: supervised learning is incorrect because it is because it requires labeled data and does not focus on finding patterns without predefined labels.

B. ✓ Correct: Unsupervised learning is correct as it aims to discover the underlying structure of data through methods like clustering or dimensionality reduction.

C. × Incorrect: reinforcement learning is incorrect because it is because it involves an agent interacting with its environment, which is unrelated to identifying unlabeled data patterns.

D. × Incorrect: deep learning is incorrect because it is because while deep neural networks can be used for various tasks including unsupervised learning, the term itself does not specify pattern identification in unlabeled data.

A. ✓ Correct: To measure how well a model can handle unexpected or adversarial inputs is correct because it accurately describes the concept of semantic robustness in evaluating AI models for their reliability and accuracy with different types of input data.

B. × Incorrect: To check if the model's responses align with ground truth data is incorrect because while important, this focuses on alignment rather than handling unexpected inputs.

C. × Incorrect: To ensure that the knowledge base retrieves highly relevant information is incorrect because retrieval relevance pertains to knowledge bases, not semantic robustness.

D. × Incorrect: To assess the correctness of retrieved texts and generated responses is incorrect because ness relates more to ground truth and accuracy in responses.

A. ✓ Correct: Operating system updates is because customers must manage guest OS updates for EC2 instances.

B. × Incorrect: Physical security of facilities is incorrect because it is because AWS manages this aspect of the infrastructure.

C. × Incorrect: AWS-provided firewall configuration is incorrect as it refers to security groups, which are managed by customers but not considered a full responsibility.

D. × Incorrect: Data encryption options is incorrect because it is because while data encryption is important, managing guest OS updates is more specific to EC2.

A. × Incorrect: Responses API is incorrect because it is because it allows submitting prompts and generating responses but does not enable server-side tool integration.

B. × Incorrect: Chat Completions API is incorrect because it is because it provides a way to communicate with the model in a conversational format, but it does not support server-side tools execution.

C. × Incorrect: Converse API is incorrect because it is because it enables conversation-based interactions with models but does not facilitate server-side tool integration.

D. ✓ Correct: AgentCore Gateway is because it supports integrating server-side tools with Amazon Bedrock models without requiring client-side orchestration.

A. ✓ Correct: Baseline dataset is correct because it is the baseline dataset is essential for establishing reference points against which current data quality can be compared.

B. × Incorrect: Training dataset is incorrect because as the training dataset is not specifically used for ongoing drift detection but rather for initial model training.

C. × Incorrect: Validation dataset is incorrect because since validation datasets are typically used during model development to tune hyperparameters and assess performance, not for continuous monitoring.

D. × Incorrect: Test dataset is incorrect because s are generally reserved for final evaluation of model accuracy and do not serve as a basis for real-time data quality checks.

A. ✓ Correct: Classification involves predicting categorical outcomes based on input data points.

B. × Incorrect: regression is incorrect because as regression predicts numerical values rather than categories.

C. × Incorrect: clustering is incorrect because groups similar objects together without predefined labels and does not predict specific categories.

D. × Incorrect: prediction is incorrect because as prediction can refer to any type of task but does not specify a categorical label output.

A. ✓ Correct: SHAP is specifically mentioned as using Shapley values for feature attributions in Amazon SageMaker Clarify.

B. × Incorrect: Partial Dependence Plots (PDPs) is incorrect because although related; PDPs show marginal effects but do not provide feature attributions based on Shapley values.

C. × Incorrect: Baseline dataset is incorrect because it is a baseline dataset is used to detect drift and monitor data quality, not for providing feature attributions.

D. × Incorrect: SageMaker Autopilot is incorrect because as SageMaker Autopilot handles preprocessing tasks like handling missing values and cross-validation resampling.

A. ✓ Correct: Automated sensitive data discovery provides continuous monitoring of S3 buckets for potential security risks such as publicly accessible buckets or sensitive data.

B. × Incorrect: S3 bucket monitoring is incorrect because while S3 bucket monitoring can be important, it does not specifically refer to the visibility and risk assessment provided by Macie.

C. × Incorrect: AWS Security Hub integration is incorrect because although AWS Security Hub integration helps in managing findings, it's not a primary method for providing initial visibility into risks as automated discovery does.

D. × Incorrect: IAM policy reviews is incorrect because it is related to access control rather than the continuous monitoring of data security risks.

A. × Incorrect: Base model is incorrect as it describes a packaged version of a foundation model, not the coordination between models and data.

B. × Incorrect: Model provider is incorrect because it is a model provider supplies or hosts models; it does not describe the application that coordinates model use with enterprise data.

C. × Incorrect: Workflow routing is incorrect because describes movement through a process, not the Bedrock term for the application that carries out coordinated tasks.

D. ✓ Correct: Agent is correct. It describes an application carrying out orchestrations involving foundation models and enterprise data.

A. ✓ Correct: The foundation model inference is discarded is correct because if the input evaluation results in a guardrail intervention, a configured blocked message response is returned and the foundation model inference is discarded.

B. × Incorrect: The response is returned without any modifications is incorrect because it is the response is only returned without modifications if it passes all evaluations successfully.

C. × Incorrect: A pre-configured blocked message is sent to the user is incorrect because but this does not imply that the inference process continues.

D. × Incorrect: The evaluation process is skipped for improved latency is incorrect because evaluation processes are not skipped; they occur in parallel for efficiency.

A. ✓ Correct: Distributed is correct because it aligns with the documentation stating that SageMaker AI provides managed ML algorithms for efficient distributed environment training.

B. × Incorrect: Batch processing is incorrect because as batch processing, while related to data processing in machine learning, is not specifically highlighted as a key feature of Amazon SageMaker AI's training options.

C. × Incorrect: Real-time inference is incorrect because it refers to model deployment rather than the training phase.

D. × Incorrect: Edge computing is incorrect because since edge computing pertains more to IoT and local device processing, which is outside the scope of Amazon SageMaker AI's primary functions.

A. × Incorrect: No additional charges is incorrect because there are always charges when a guardrail evaluates and potentially blocks an input or response.

B. × Incorrect: Charges for the model inference only is incorrect because as blocking an input does not necessarily result in a model inference charge; it depends on whether the model was invoked before intervention.

C. × Incorrect: Charges for the guardrail evaluation and the model response is incorrect because since this would apply if both prompt and response were blocked, but here we're considering only when the initial prompt is blocked without further inference.

D. ✓ Correct: Charges for the guardrail evaluation of the prompt is correct because charges are incurred for evaluating the input prompt even if no further actions (like generating a response) occur.

A. × Incorrect: They are stored for auditing purposes is incorrect because amazon Bedrock does not store or log prompts and completions for auditing purposes.

B. × Incorrect: They are used to train AWS models is incorrect because amazon Bedrock does not use customer interactions to train AWS models.

C. ✓ Correct: They are not logged or distributed to third parties is correct because amazon Bedrock maintains the confidentiality of customer data by not logging or sharing it with third parties.

D. × Incorrect: They are shared with model providers is incorrect because model providers do not have access to deployment accounts where customer data resides.

A. ✓ Correct: To enhance model responses with data from proprietary sources is correct because it directly aligns with the purpose of integrating data sources for retrieval workflows in Amazon Bedrock.

B. × Incorrect: To train foundation models on diverse datasets is incorrect because training foundation models on diverse datasets is not specific to knowledge bases; this is a general practice in machine learning.

C. × Incorrect: To generate natural language queries for unstructured databases is incorrect because generating natural language queries for unstructured databases is part of querying structured data stores, but it's not the primary purpose of knowledge bases.

D. × Incorrect: To convert images into text is incorrect because converting images into text is related to services like Amazon Textract and is unrelated to retrieval workflows with knowledge bases.

A. × Incorrect: AWS Management Console is incorrect because while the AWS Management Console can be used, it's not specifically designed for enhancing software development workflows within an integrated environment.

B. × Incorrect: Microsoft Teams chat application is incorrect because but relevant; Amazon Q Developer can indeed be used in chat applications like Microsoft Teams to provide assistance outside of coding environments.

C. ✓ Correct: Integrated Development Environment (IDE) is correct because as Amazon Q Developer integrates with various IDEs such as Visual Studio Code and JetBrains, providing features that enhance the coding experience directly within these development tools.

D. × Incorrect: Amazon SageMaker is incorrect because while Amazon SageMaker is a powerful tool for machine learning model deployment, it does not describe an environment where Amazon Q Developer can be used to assist software developers.

A. ✓ Correct: Amazon Transcribe provides real-time speech-to-text transcription services, making it ideal for converting spoken words into text as they occur.

B. × Incorrect: Amazon Translate is incorrect because specializes in translating text from one language to another and does not handle speech-to-text conversion.

C. × Incorrect: Amazon Polly is incorrect because converts text into lifelike speech and is used primarily for text-to-speech applications, not the reverse process of converting speech to text.

D. × Incorrect: Amazon Lex is incorrect because builds conversational interfaces and is designed for building chatbots and voice-based interactions but does not provide real-time transcription services.

A. ✓ Correct: Transparency is correct because it directly supports the objective of ensuring that stakeholders understand and can make informed decisions about their engagement with an AI system.

B. × Incorrect: Explainability is incorrect because it focuses on understanding how AI makes decisions, not informing stakeholders about their engagement choices.

C. × Incorrect: Privacy and security is incorrect because they are concerned with protecting data and models from exfiltration and adversarial inputs, not stakeholder engagement.

D. × Incorrect: Safety is incorrect because it involves reducing harmful system output and misuse, rather than enabling informed decision-making by stakeholders.

A. ✓ Correct: customModelKmsKeyId is correct because as it specifies the KMS key ID used to encrypt the custom model.

B. × Incorrect: customerEncryptionKeyArn is incorrect because it is used for agents, not models.

C. × Incorrect: kmsKeyArn is incorrect because it is used for data source ingestion jobs and vector stores, not custom models.

D. × Incorrect: modelCustomizationJobId is incorrect because identifies a job but does not specify encryption keys.

A. × Incorrect: Validation dataset is incorrect because it is because while it can be used, a training dataset is required before starting any customization job.

B. ✓ Correct: Training dataset is because Amazon Bedrock requires at least a prepared training dataset to begin customizing models. It forms the basis of learning for the model.

C. × Incorrect: Knowledge base is incorrect as it enhances responses with data from proprietary sources but is not essential for initiating a model customization job.

D. × Incorrect: Prompt templates is incorrect because it is used in agents to collect additional information through natural conversation and do not serve as datasets for training.

A. ✓ Correct: To generate comprehensive responses to natural language queries from users is correct because amazon Q Business generates comprehensive responses to user queries by analyzing information across all accessible enterprise content.

B. × Incorrect: To provide real-time translation services for multi-lingual teams is incorrect because real-time translation services are provided by other AWS services like Amazon Translate, not Amazon Q Business.

C. × Incorrect: To manage and monitor machine learning models deployed on AWS is incorrect because as managing and monitoring machine learning models is handled by services such as Amazon SageMaker Model Monitor, not Amazon Q Business.

D. × Incorrect: To facilitate secure data sharing between third-party applications is incorrect because since secure data sharing between third-party applications is facilitated through plugins and integrations but not the primary purpose of Amazon Q Business.

A. × Incorrect: Ready-to-use models is incorrect because it refer to pre-built solutions for specific use cases, such as image or text analysis.

B. ✓ Correct: Canvas chat leverages open-source and Amazon LLMs to assist with tasks like generating content and summarizing documents.

C. × Incorrect: Custom model training is incorrect because it involves building a predictive model tailored to your data and use case.

D. × Incorrect: Data import is incorrect because it refers to the process of bringing in datasets for analysis or model training.

A. ✓ Correct: Feature attributions is correct because sageMaker Clarify provides feature attributions which help data scientists understand how different features impact model predictions, aiding in debugging and improving models.

B. × Incorrect: Bias metrics is incorrect because while bias metrics are important for detecting unfairness, they do not directly assist in the process of debugging or enhancing model performance.

C. × Incorrect: Model governance reports is incorrect because it is useful for compliance and regulatory purposes but do not provide direct insights into how to improve a model's predictive accuracy.

D. × Incorrect: Inference model drift alerts is incorrect because it monitor changes in model behavior over time but do not offer specific guidance on improving the model itself.

A. ✓ Correct: Amazon Athena provides a way to run SQL queries on CloudTrail logs stored in S3.

B. × Incorrect: AWS Lambda is incorrect because as AWS Lambda is for running code without provisioning or managing servers, not analyzing logs.

C. × Incorrect: Amazon RDS is incorrect because since Amazon RDS is for relational database services and not for querying log data.

D. × Incorrect: Amazon VPC is incorrect because it is for virtual networking and does not support querying CloudTrail logs.

A. ✓ Correct: Supervised fine-tuning involves providing labeled data to train a model for specific tasks, improving its performance and creating better customer experiences.

B. × Incorrect: Reinforcement fine-tuning is incorrect because it uses feedback-based learning through reward functions rather than labeled input-output pairs.

C. × Incorrect: Distillation is incorrect because transfers knowledge from a larger teacher model to a smaller student model using diverse high-quality responses generated by the teacher model.

D. × Incorrect: Context description is incorrect because it provides context to tasks in prompt engineering, not for training models with supervised data.