AWS Certified AI Practitioner (AIF-C01) Practice Test – 247 Free Exam Questions with Answers

Correct answer: C. This is known as a hallucination, where the model generates seemingly accurate information that is, in fact, incorrect or fabricated

When an LLM generates plausible but factually incorrect information, it is called a hallucination. Terms like overfitting or data drift describe model training or data distribution issues, not the specific behavior of generating fabricated text.

Correct answer: B. Influences the percentage of most-likely candidates that the model considers for the next token

The Top P parameter defines the percentage of the most likely candidates the model considers for the next token. Temperature controls creativity, while Top K limits the fixed number of candidates considered.

Correct answer: B. With Amazon Bedrock, you can choose the underlying Foundation Model. However, Amazon Q does not allow you to choose the underlying Foundation Model · C. Amazon Q is a generative AI–powered assistant that allows you to create pre-packaged generative AI applications, whereas, Amazon Bedrock provides an environment to build and scale generative AI applications using a Foundation Model (FM)

Amazon Bedrock is a managed service for building generative AI applications using a choice of foundation models. Amazon Q is a pre-packaged assistant without model selection. Remember that Bedrock exposes models via API, while Q provides a ready-to-use assistant.

Correct answer: D. Amazon SageMaker Ground Truth

Amazon SageMaker Ground Truth creates high-quality labeled datasets using human workers. Feature Store manages ML features, while JumpStart deploys pre-trained models. Ground Truth is essential for data labeling.

Correct answer: B. Utilize a Retrieval-Augmented Generation (RAG) system by indexing all product catalog PDFs and configuring the LLM chatbot to reference this system for answering queries

A Retrieval-Augmented Generation system indexes documents to fetch relevant context per query. Fine-tuning is expensive and slow to update, and passing all PDFs exceeds context windows and inflates costs.

Correct answer: A. Amazon Bedrock · B. Amazon SageMaker JumpStart

Amazon Bedrock and SageMaker JumpStart provide managed environments to deploy and customize large language models. AWS Trainium and Inferentia are hardware chips, while Amazon Q is an assistant.

Correct answer: D. The company should use BLEU (Bilingual Evaluation Understudy) score, a metric specifically designed to evaluate the quality of text that has been machine-translated by comparing it with one or more reference translations

The BLEU score evaluates machine translation quality by comparing n-gram overlaps with reference translations. ROUGE is primarily used for summarization, while BLEU remains the standard for translation.

Correct answer: C. Shapley values provide a local explanation by quantifying the contribution of each feature to the prediction for a specific instance, while PDP provides a global explanation by showing the marginal effect of a feature on the model's predictions across the dataset. Use Shapley values to explain individual predictions and PDP to understand the model's behavior at a dataset level

Shapley values offer local explainability by assigning contribution scores to features for a specific prediction. Partial Dependence Plots provide global explainability by showing feature effects across the dataset.

Correct answer: B. Amazon Kendra

Amazon Kendra is an enterprise search service powered by machine learning that connects to various data repositories to provide accurate answers. Textract extracts text from documents, while Comprehend performs natural language processing but neither functions as a unified search endpoint.

Correct answer: A. The company should mask the confidential information from the model responses by leveraging Amazon Bedrock Guardrails

Amazon Bedrock Guardrails can dynamically scan and mask sensitive data or personally identifiable information in model responses. Retraining the model without the data is inefficient and expensive, while encryption only protects data in transit or at rest, not in the generated output.

Correct answer: B. The company should opt for on-demand pricing, which allows it to pay only for the actual usage of resources without any long-term commitments

On-demand pricing is best for companies with unpredictable usage because you only pay for what you use without long-term commitments. Spot and Reserved Instances are EC2 pricing models, while Provisioned Throughput requires a predictable usage commitment.

Correct answer: B. Amazon SageMaker Clarify

Amazon SageMaker Clarify provides feature attribution tools to explain how specific input features contribute to a model's predictions. Feature Store manages data inputs, Model Monitor tracks production drift, and Canvas is a no-code interface.

Correct answer: C. Tokens

Tokens represent the discrete units of text processed by generative AI models. Embeddings and vectors are numerical representations of these tokens, while the context window measures the total token capacity the model can process.

Correct answer: C. Using generative AI to create photorealistic images from textual descriptions

Creating photorealistic images from text descriptions is a classic generative AI capability because it produces entirely new content. The distractors fail because predicting housing prices or classifying images rely on traditional supervised machine learning rather than generative models.

Correct answer: C. AWS is responsible for the security "of" the cloud, including infrastructure, hardware, and software, while the customer is responsible for security "in" the cloud, including data, applications, and access management

AWS operates the security of the cloud, meaning it handles the underlying infrastructure, hardware, and facilities. Customers handle security in the cloud, which includes securing their data, managing access, and configuring applications.

Correct answer: D. Predict product demand to accurately vary inventory and pricing at different store locations

Amazon Forecast uses time-series data to predict future metrics, making it ideal for retail demand planning and inventory optimization. The other options map to different AI services like Personalize for recommendations or Lex for conversational interfaces.

Correct answer: A. Amazon Polly is used to deploy high-quality, natural-sounding human voices in dozens of languages · B. Amazon Comprehend service uses machine learning to find insights and relationships in the text

Amazon Polly converts text into lifelike speech, and Amazon Comprehend uses natural language processing to extract insights from text. The distractors mismatch the services, such as pairing Transcribe with conversational interfaces instead of Lex.

Correct answer: B. FMs use self-supervised learning to create labels from input data, however, fine-tuning an FM is a supervised learning process

Foundation models rely on self-supervised learning to generate implicit labels from vast amounts of raw data. Fine-tuning customizes the model for specific tasks using labeled examples, which makes it a supervised learning process.

Correct answer: C. OpenSearch Serverless vector store

Knowledge Bases for Amazon Bedrock natively defaults to OpenSearch Serverless for vector storage. Although it supports Pinecone and Aurora, the default store is automatically created if you do not provide one.

Correct answer: A. Amazon Transcribe and Amazon Comprehend

Amazon Transcribe converts audio calls into text. Amazon Comprehend then runs sentiment analysis on that text. Avoid Rekognition for audio, as it handles computer vision tasks.

Correct answer: C. The company should use hyperparameters for model tuning, which involves adjusting parameters such as regularization, learning rates, and dropout rates to enhance the model's ability to generalize well to new data

Hyperparameter tuning adjusts regularization and dropout rates to prevent overfitting, improving model generalization. While adding data helps, hyperparameter tuning specifically targets the gap between training and production.

Correct answer: D. Augment the data by generating new instances of data for underrepresented groups

Data augmentation directly balances underrepresented groups by creating new synthetic training examples, which mitigates bias. Manual correction is not scalable, regularization only prevents overfitting, and switching models ignores the root cause of bad data.

Correct answer: A. Amazon Rekognition

Amazon Rekognition offers pre-trained computer vision capabilities that you can customize for specific business needs. Textract is strictly for text extraction, while SageMaker is a broader service requiring you to build models from scratch.

Correct answer: C. Negative prompting

Negative prompting tells a generative AI model exactly what concepts or terms to exclude from its output. This is ideal for blocking competitor names or sensitive topics, whereas few-shot or chain-of-thought prompt techniques are used to guide desired content.

Correct answer: A. Foundation Models serve as a broad base for various AI applications by providing generalized capabilities, whereas Large Language Models are specialized for understanding and generating human language

Foundation Models are broad, general-purpose systems trained on massive datasets for various modalities like text or images. Large Language Models are a specialized subset of this category focused exclusively on understanding and generating human language.

Correct answer: D. Instructions, Context, Input data, Output Indicator

A well-crafted prompt consists of instructions, context, input data, and an output indicator. Remember that hyperparameters and parameters are configured at the model level, not manually embedded directly within the prompt text itself.

Correct answer: C. The data should include user-input along with the correct user intent, providing examples of user queries and the corresponding intent

Few-shot prompting teaches the model by pairing user inputs with their corresponding correct intent. Providing matching input and response pairs might show the final answer, but it fails to explicitly demonstrate the intermediate intent classification step.

Correct answer: A. The company should enable model invocation logging, which allows for detailed logging of all requests and responses during model invocations in Amazon Bedrock

Enabling model invocation logging in Amazon Bedrock captures the actual prompt inputs and generated outputs for auditing. CloudTrail only tracks control plane API calls for account security and does not record the payload data of the model responses.

Correct answer: B. Confusion matrix

A confusion matrix evaluates classification models by showing true positives, true negatives, false positives, and false negatives. Mean Absolute Error and Root Mean Squared Error are regression metrics designed for continuous numerical predictions instead of categorical ones.

Correct answer: A. Reinforcement learning involves an agent interacting with an environment by taking actions and receiving rewards or penalties, learning a policy to maximize cumulative rewards over time

Reinforcement learning uses an agent that interacts with an environment, receiving rewards or penalties to maximize cumulative success. It does not rely on labeled datasets like supervised learning or group unlabeled data like unsupervised learning.

Correct answer: C. The company should use Domain Adaptation Fine-Tuning, which involves fine-tuning the model on domain-specific data to adapt its knowledge to that particular domain · D. The company should use Continued Pre-Training, which involves further training the model on a large corpus of domain-specific data, enhancing its ability to understand domain-specific terms, jargon, and context

Domain Adaptation Fine-Tuning and Continued Pre-Training are the correct approaches because they use large amounts of unlabeled or labeled domain-specific data to teach the model specialized terminology and patterns. On the exam, remember that reinforcement learning relies on human feedback rather than raw domain knowledge.

Correct answer: D. Exploratory Data Analysis (EDA)

Exploratory Data Analysis is correct because calculating statistical summaries and generating visualizations to uncover trends is the core focus of EDA. Data preparation is the strongest distractor, but it involves cleaning and transforming data rather than exploring it for initial insights.

Correct answer: A. The company should use Amazon OpenSearch Service, which is designed to provide fast search capabilities and supports full-text search, indexing, and similarity scoring

Amazon OpenSearch Service is the correct choice because it provides native support for vector searches, fast index lookups, and similarity scoring required for retrieval-augmented generation workloads. Relational databases like Aurora lack these specialized semantic search capabilities.

Correct answer: A. Top K

Top K is the correct parameter because it directly limits the token pool to a specific number of the most likely next candidates. Temperature controls randomness globally, while Top P uses a probability threshold instead of a fixed count.

Correct answer: D. Amazon SageMaker Model Dashboard

The SageMaker Model Dashboard provides a centralized portal to track deployed models, monitor endpoints, and troubleshoot real-time inference issues. JumpStart is for deploying pre-trained models, while Clarify focuses solely on detecting bias in your data.

Correct answer: A. Sampling bias

Sampling bias is correct because the skewed outcomes stem directly from certain demographic groups being overrepresented or underrepresented in the training data. Measurement bias involves faulty sensors, while confirmation bias applies to human expectations.

Correct answer: C. Enable multilingual user experiences in your applications

Enabling multilingual experiences is the correct answer because translation tasks belong to Amazon Translate, not Rekognition. Rekognition is strictly for computer vision tasks like facial recognition, celebrity detection, and searching image libraries.

Correct answer: B. The company should use Provisioned Throughput mode, which allows the company to reserve a specific amount of capacity in advance

Provisioned Throughput is correct because Amazon Bedrock requires you to purchase this reserved capacity to test and deploy any models customized via fine-tuning. On-Demand mode only applies to base foundation models and cannot access your customizations.

Correct answer: B. Use higher Temperature to get more creative responses for the same prompt on Amazon Bedrock

A higher Temperature value is the correct setting because it increases the randomness of token selection, yielding more creative and diverse responses. Lower temperature values force the model to produce deterministic and predictable outputs.

Correct answer: D. Generative AI encompasses models and algorithms capable of creating new content such as text, images, and audio based on patterns learned from existing data

Generative AI creates new content like text, images, and audio by learning patterns from existing data. Options focusing on data retrieval, predefined tasks, or just analyzing data without creating content are incorrect because they fail to produce original output.

Correct answer: C. The company should reduce the number of tokens in the input

Generative AI model costs on Amazon Bedrock are calculated based on the number of input and output tokens processed. Adjusting temperature or top-P affects the randomness and diversity of the output but does not lower the usage price.

Correct answer: D. You can evaluate, compare, and select Foundation Models quickly based on pre-defined quality and responsibility metrics · E. Pre-trained models are fully customizable for your use case with your data

SageMaker JumpStart allows users to quickly evaluate foundation models and deploy customizable pre-trained models. SageMaker Canvas is the AWS no-code solution, while training data is never used to update the base models surfaced in JumpStart.

Correct answer: B. Model evaluation is the process of evaluating and comparing model outputs to determine the model that is best suited for a use case, whereas, model inference is the process of a model generating an output (response) from a given input (prompt)

Model inference is the act of a model generating an output from a given prompt, whereas model evaluation is the process of comparing these outputs to find the best model. The other options incorrectly swap or merge these two distinct concepts.

Correct answer: A. Manage machine learning experiments

MLflow integrates with SageMaker to help data scientists track, organize, and compare iterative machine learning experiments. Automatic tuning is handled by SageMaker AMT, while Canvas provides no-code tools and Ground Truth handles data labeling.

Correct answer: D. Transformer models use a self-attention mechanism and implement contextual embeddings

Transformer models use self-attention mechanisms to weigh the importance of words in a sequence alongside contextual embeddings. The other options describe generative adversarial networks, variational autoencoders, and diffusion models instead of transformers.

Correct answer: C. Bidirectional Encoder Representations from Transformers (BERT)

BERT uses bidirectional context to generate dynamic word embeddings, capturing nuanced meanings based on surrounding text. Older models like Word2Vec rely on static embeddings, while SVD and PCA are dimensionality reduction techniques that lack language context.

Correct answer: A. Knowledge Bases for Amazon Bedrock

Knowledge Bases for Amazon Bedrock provides a fully managed end-to-end Retrieval Augmented Generation workflow by ingesting documents into a vector store. Guardrails filter harmful content, while continued pretraining customizes model weights without retrieving external data.

Correct answer: B. Agents for Amazon Bedrock

Agents for Amazon Bedrock autonomously orchestrate multi-step tasks by integrating foundation models with proprietary data and APIs. Knowledge Bases only retrieve information, whereas Guardrails and watermark detection focus on safety and image provenance respectively.

Correct answer: D. Continued pre-training uses unlabeled data to pre-train a model, whereas, fine-tuning uses labeled data to train a model

Continued pre-training uses unlabeled data to adapt a foundation model to new domains, while fine-tuning uses labeled data to optimize it for specific tasks. Expect exam questions testing the distinction between unsupervised domain adaptation and supervised task-specific training.

Correct answer: B. Define the use case of the application narrowly

Defining the use case narrowly gives the research team specific, bounded requirements, which directly drives correct model selection. Broadly defining an audience or listing data sources are preliminary discovery steps, but they do not by themselves ensure the best model is chosen.

Correct answer: C. Decision Trees

Decision trees provide high interpretability through clear, visual rules, making them the best fit when you must explain how features drive a classification. Neural networks act as black boxes, and while powerful, they lack the inherent transparency required here.

Correct answer: A. Elevate the customer experience with ML-powered personalization

Amazon Personalize uses machine learning to generate real-time, customized product and content recommendations. The other options describe different AWS AI services: Comprehend for text insights, Transcribe for speech-to-text, and Polly for text-to-speech.

Correct answer: D. Plagiarism

Detecting unauthorized AI-generated essays addresses plagiarism by ensuring applicants submit original work. Hallucination refers to an AI inventing false facts, while bias involves unfair discrimination, neither of which is the primary concern here.

Correct answer: A. Generative AI can automate the creation of new data based on existing patterns, enhancing productivity and innovation

Automating the creation of new data based on existing patterns is the primary advantage of generative artificial intelligence, directly enhancing productivity and driving innovation. The other options are incorrect because no AI can fully replace humans or guarantee absolute security.

Correct answer: B. Large Language Model (LLM) · C. Retrieval-Augmented Generation (RAG)

Amazon Q Business uses Large Language Models to generate text and Retrieval-Augmented Generation to fetch context from connected enterprise data. Diffusion models, GANs, and VAEs primarily generate images or synthetic tabular data, not enterprise search answers.

Correct answer: C. Amazon Comprehend · D. Amazon Bedrock

Amazon Comprehend directly analyzes text sentiment, while Amazon Bedrock allows foundation models to perform custom sentiment analysis. Rekognition handles images, Textract extracts document text, and Personalize builds recommendation systems, so none of those process written sentiment.

Correct answer: C. Asynchronous inference

Asynchronous inference is specifically designed for payloads under one gigabyte where immediate real-time responses are unnecessary. Batch transform is wrong because it targets massive datasets rather than these smaller daily payloads processed individually.

Correct answer: A. The company should leverage reinforcement learning (RL), where rewards are generated from positive customer feedback to train the chatbot in optimizing its responses

Reinforcement learning uses reward signals from positive feedback to dynamically optimize chatbot responses over time. Incremental training could technically achieve ongoing adaptation, but it lacks the explicit reward optimization focus that reinforcement learning provides.

Correct answer: A. This concept is referred to as a context window, which determines the amount of text or information the model can consider at once while generating a response, typically measured in tokens rather than characters

The context window dictates the maximum amount of text, measured in tokens, that a model can process in a single interaction. Tokens themselves are merely the individual subword units, while embeddings represent text as numerical vectors for semantic understanding.

Correct answer: D. AWS Trusted Advisor

AWS Trusted Advisor provides real-time guidance to help you follow AWS best practices across cost, performance, security, and fault tolerance. Distractors like AWS Config focus on resource configuration tracking and compliance auditing rather than holistic optimization recommendations.

Correct answer: D. The company should use transfer learning, a method where a model pre-trained on one task is adapted to improve performance on a different but related task by leveraging knowledge from the original task

Transfer learning applies knowledge from a pre-trained model to a related task. While federated learning is a better fit for the scenario of distributed models sharing insights, transfer learning is the intended answer among the provided choices.

Correct answer: C. Fine-tuning changes the weights of the FM whereas Retrieval-augmented generation (RAG) does not change the weights of the FM

Fine-tuning alters the underlying weights of a foundation model through additional training. Retrieval-augmented generation, or RAG, leaves model weights unchanged and instead grounds responses using external knowledge sources.

Correct answer: B. Amazon SageMaker Feature Store

Amazon SageMaker Feature Store is a purpose-built repository designed to store, share, and manage features for machine learning models. Data Wrangler is used for data preparation, Clarify detects bias, and Ground Truth handles data labeling.

Correct answer: B. Feature Engineering involves selecting, modifying, or creating features from raw data to improve the performance of machine learning models, and it is important because it can significantly enhance model accuracy and efficiency

Feature engineering involves selecting, modifying, or creating features from raw data to significantly improve machine learning model accuracy. Distractors incorrectly confuse feature engineering with data collection, visualization, or hyperparameter tuning.

Correct answer: A. Amazon SageMaker Model Monitor · C. Amazon SageMaker Clarify

SageMaker Clarify detects bias and explains predictions to ensure fairness, while SageMaker Model Monitor tracks production models for drift. The other services are general security or compliance tools, not specifically responsible AI practices.

Correct answer: C. Chain-of-thought prompting

Chain-of-thought prompting breaks complex problems into smaller, logical intermediate steps to improve model reasoning capabilities. Zero-shot and few-shot prompting provide examples rather than explicitly decomposing the reasoning process.

Correct answer: A. Amazon SageMaker Model Monitor · E. Amazon Augmented AI (Amazon A2I)

SageMaker Model Monitor tracks deployed models for data drift, while Amazon A2I provides human review workflows for low-confidence predictions. Ground Truth handles training data labeling, not production model oversight.

Correct answer: B. A token is a sequence of characters that a model can interpret or predict as a single unit of meaning, whereas, an embedding is a vector of numerical values that represents condensed information obtained by transforming input into that vector

A token is a sequence of characters processed as a single unit of meaning, whereas an embedding represents that input as a numerical vector. Embeddings condense information into mathematical space for similarity comparisons.

Correct answer: C. Foundation models can perform a wide range of tasks across different domains by leveraging their extensive pre-training on large datasets

Foundation models leverage extensive pre-training on massive datasets to generalize across multiple domains and perform diverse tasks. They are not restricted to single tasks or solely structured data like legacy machine learning systems.

Correct answer: B. Top P

Top P defines the percentage of most likely candidates considered for the next token, controlling the output pool dynamically. Top K sets a strict number of candidates, while Temperature controls overall randomness.

Correct answer: C. The model can recreate new animal images that were not in the training dataset

Generative models learn the underlying distribution of training data to synthesize entirely new, unseen examples. Discriminative models, like classifiers, only map inputs to specific labels. Use the term recreate to spot the generative choice.

Correct answer: A. Image processing focuses on enhancing and manipulating images for visual quality, whereas computer vision involves interpreting and understanding the content of images to make decisions

Image processing handles pixel manipulations like filtering and compression, whereas computer vision interprets image content to make automated decisions. Remember that vision systems aim to mimic human understanding, not just alter pixels.

Correct answer: D. Model training in deep learning involves using large datasets to adjust the weights and biases of a neural network through multiple iterations, using techniques such as gradient descent to minimize the error

Deep learning training iteratively adjusts neural network weights and biases using large datasets and optimization techniques like gradient descent to minimize error. Network weights are never set manually using predefined rules.

Correct answer: C. They facilitate easier debugging and optimization · D. They foster trust and confidence in model predictions

Transparent models facilitate easier debugging by revealing how inputs become outputs and foster trust among stakeholders in high-stakes industries. Computational power and integration complexity depend on architecture, not model transparency.

Correct answer: A. Neural networks consist of layers of nodes (neurons) that process input data, adjusting the weights of connections between nodes through training to recognize patterns and make predictions

Neural networks rely on layers of interconnected nodes that process inputs and adjust connection weights during training to recognize patterns. They learn directly from data rather than relying on explicit programming or storing outcomes.

Correct answer: A. Applying multiple layers of security measures including input validation, access controls, and continuous monitoring to address vulnerabilities

Defense-in-depth relies on applying multiple overlapping security measures like input validation, strict access controls, and continuous monitoring. Relying on a single security layer, such as just encryption, leaves the application vulnerable.

Correct answer: B. Amazon SageMaker Model Dashboard

Amazon SageMaker Model Dashboard provides a centralized visual interface to track, manage, and monitor all machine learning models across your AWS account. Model Monitor only checks production data quality without providing a holistic model inventory.

Correct answer: B. Machine Learning models can be deterministic or probabilistic or a mix of both

Machine learning models can be deterministic, probabilistic, or a combination of both depending on their specific architecture and design. Associating deterministic behavior only with supervised learning is a common distractor.

Correct answer: C. By using techniques such as cross-validation, regularization, and pruning to simplify the model and improve its generalization

Applying techniques like cross-validation, regularization, and pruning simplifies models to ensure they generalize well to unseen data. Increasing model complexity only exacerbates overfitting by memorizing noise.

Correct answer: C. The bias versus variance trade-off refers to the challenge of balancing the error due to the model's complexity (variance) and the error due to incorrect assumptions in the model (bias), where high bias can cause underfitting and high variance can cause overfitting

High bias stems from overly simplistic assumptions causing underfitting, while high variance makes a model overly sensitive and prone to overfitting. Remember that reversing these definitions is a common trap on exam questions.

Correct answer: D. Response length

The response length parameter dictates the maximum number of tokens a model generates. Stop sequences halt generation based on specific text patterns, while Top P and Top K control randomness rather than length.

Correct answer: B. Temperature

Temperature controls the randomness or creativity of a language model's responses. Top P and Top K narrow the token pool for deterministic outputs, while stop sequences simply terminate text generation early.

Correct answer: D. Overfitting occurs when a model performs well on the training data but poorly on new, unseen data, while underfitting occurs when a model performs poorly on both the training data and new, unseen data

Overfitting happens when a model memorizes training data, performing poorly on unseen data. Underfitting occurs when a model is too simple to capture underlying patterns, performing poorly on both training and unseen data.

Correct answer: B. Amazon Q in Connect

Amazon Q in Connect is the correct choice because it uses real-time customer conversations and company content to suggest actions and responses for contact center agents. Amazon Q Business targets enterprise data assistants, while Amazon Q Developer focuses on coding tasks.

Correct answer: A. Guardrails for Amazon Bedrock · C. Model Evaluation on Amazon Bedrock

Model Evaluation on Amazon Bedrock helps select the best foundation model for specific use cases. Guardrails for Amazon Bedrock implements safety controls to filter harmful content. Other options like Comprehend or SageMaker Clarify do not directly select models or moderate LLM outputs.

Correct answer: D. Amazon SageMaker Data Wrangler

Amazon SageMaker Data Wrangler provides a single visual interface for data selection, cleansing, exploration, and visualization. Feature Store only manages features, while Clarify detects bias, and neither offers the complete data preparation workflow shown here.

Correct answer: C. Amazon SageMaker Canvas

Amazon SageMaker Canvas provides a visual point-and-click interface allowing business analysts to build ML models without coding. Data Wrangler is only for data preparation, whereas Canvas handles the complete workflow from data to deployment.

Correct answer: B. Data access control involves authentication and authorization of users, whereas data integrity ensures the data is accurate, consistent, and unaltered

Data access control handles authentication and authorization to manage user permissions. Data integrity ensures the information remains accurate and unaltered during storage or processing. Distractors fail by confusing encryption tasks with permission management.

Correct answer: A. Model parameters are values that define a model and its behavior in interpreting input and generating responses. Hyperparameters are values that can be adjusted for model customization to control the training process

Model parameters are internal values learned during training that define model behavior, like weights. Hyperparameters are external settings adjusted before training to control the process, like learning rate. The remaining options incorrectly swap these definitions.

Correct answer: D. Difficulty in collecting and preparing high-quality data for training models

Data collection and preparation are widely recognized as the most time-consuming bottlenecks in machine learning, making high-quality data the primary challenge. The distractors describing a lack of algorithms or use cases are factually incorrect for modern machine learning.

Correct answer: C. Smaller models are cheaper to use than larger models · E. You can use a customized model in the Provisioned Throughput or On-Demand mode

Smaller generative models are faster and cheaper to run than larger ones, and Bedrock customized models support both On-Demand and Provisioned Throughput modes. On-Demand pricing does not require time-based term commitments.

Correct answer: D. FMs use unlabeled training data sets for self-supervised learning

Foundation models are trained on vast amounts of unlabeled data using self-supervised learning techniques to predict missing parts of the input. Supervised learning requires labeled outputs, making those options incorrect.

Correct answer: D. Benefit from massive economies of scale · E. Trade capital expense for variable expense · F. Go global in minutes and deploy applications in multiple regions around the world with just a few clicks

Cloud computing provides massive economies of scale, trades capital expenses for variable expenses, and enables rapid global deployment. On-premises infrastructure forces capacity guessing and heavy upfront data center investments.

Correct answer: B. Sentiment analysis · E. Fraud identification

Semi-supervised learning leverages a small amount of labeled data alongside a large volume of unlabeled data, which fits applications like sentiment analysis and fraud identification. Clustering and dimensionality reduction are strictly unsupervised.

Correct answer: B. Infrastructure as a Service (IaaS)

Amazon EC2 provides virtualized compute resources, categorizing it as Infrastructure as a Service because you manage the operating system and application. PaaS offerings, like Elastic Beanstalk, handle the underlying infrastructure automatically.

Correct answer: B. When temperature is set to 1

Setting the temperature to its maximum value randomizes word selection, increasing the probability of hallucinations in generative text models. A temperature of zero forces deterministic responses, effectively reducing hallucination risks.

Correct answer: D. Cloud computing refers to the on-demand delivery of IT resources and applications via the internet with pay-as-you-go pricing

Cloud computing is defined as the on-demand delivery of IT resources over the internet with pay-as-you-go pricing. For the exam, associate this model with flexibility and scalability instead of owning physical hardware.

Correct answer: A. Reinforcement learning focuses on an agent learning optimal actions through interactions with the environment and feedback, while supervised learning involves training models on labeled data to make predictions

Reinforcement learning uses an agent interacting with an environment to maximize cumulative rewards. Supervised learning, however, maps inputs to outputs using labeled datasets to make accurate predictions.

Correct answer: D. IAM Identity Center

IAM Identity Center is the recommended service to manage workforce access to Amazon Q Business applications. Standard IAM users are better suited for administrative AWS resource access rather than end-user application access.

Correct answer: A. Feature engineering for structured data often involves tasks such as normalization and handling missing values, while for unstructured data, it involves tasks such as tokenization and vectorization

Structured data feature engineering usually involves normalization and handling missing values. Unstructured data requires techniques like tokenization and vectorization to convert text or images into usable numerical formats.

Correct answer: B. AWS Audit Manager

AWS Audit Manager continuously audits your AWS usage by automating evidence collection to simplify risk assessments. AWS Artifact simply provides access to compliance reports and does not actively collect evidence.

Correct answer: C. Get answers to your AWS account-specific cost-related questions using natural language · E. Understand and manage your cloud infrastructure on AWS

Amazon Q Developer answers natural language questions about AWS costs and helps manage infrastructure. It cannot visualize cost data or autonomously deploy and modify resources, making those distractors incorrect.

Correct answer: D. The company should use a multi-modal embedding model, which is designed to represent and align different types of data (such as text and images) in a shared embedding space, allowing the chatbot to understand and interpret both forms of input simultaneously

A multimodal embedding model aligns text and images within a shared vector space for cost-effective search and querying. A generative model is overkill, more expensive, and better suited for creating new content.

Correct answer: A. A multimodal model can accept a mix of input types such as audio/text and create a mix of output types such as video/image

Multimodal models accept diverse inputs like text and audio while generating mixed outputs like images and video. The defining characteristic is handling multiple data formats simultaneously rather than being restricted to a single type.

Correct answer: A. A hiring algorithm consistently prefers candidates from a particular gender, even though the candidates' qualifications are similar across genders

Algorithmic bias occurs when a model systematically favors specific groups despite similar qualifications. Human judgment errors or random weather fluctuations fail to represent systematic machine learning bias.

Correct answer: B. SageMaker model cards include information about the model such as intended use and risk rating of a model, training details and metrics, evaluation results, and observations. AI service cards provide transparency about AWS AI services' intended use, limitations, and potential impacts

SageMaker model cards document custom model details like training metrics and risk ratings. AI service cards provide transparency for pre-trained AWS services regarding their intended use and limitations.

Correct answer: D. The company should iteratively test and adjust the chatbot prompts to ensure that its outputs consistently reflect the company's tone and style

Prompt engineering iteratively refines responses to match a specific tone or style. Lowering temperature reduces randomness but does not guarantee the desired tone, while batch inferencing and token limits are unrelated.

Correct answer: A. The company should use an optimized small language model (SLM) deployed directly on the edge device, allowing for real-time, low-latency inference

Deploying an optimized small language model directly on the edge device ensures real-time performance without network latency. Large models exceed edge hardware limits, while central APIs introduce unacceptable network delays.

Correct answer: A. ChatGPT

ChatGPT relies on the transformer architecture, utilizing self-attention mechanisms for natural language tasks. Stable Diffusion and DALL-E are diffusion models primarily used for image generation rather than text.

Correct answer: D. Artificial Intelligence > Machine Learning > Deep Learning > Generative AI

Artificial intelligence is the broadest category, containing machine learning, which contains deep learning, which enables generative AI. Knowing this nested subset hierarchy is critical for mapping technologies correctly.

Correct answer: D. Self-supervised learning

Foundation models use self-supervised learning to create labels from input data. Supervised learning requires pre-labeled data sets, whereas reinforcement learning uses reward functions rather than autonomously generating labels.

Correct answer: B. Use Knowledge Bases for Amazon Bedrock to supplement contextual information from the company's private data to the FM using Retrieval Augmented Generation (RAG)

Knowledge Bases for Amazon Bedrock uses Retrieval Augmented Generation to securely inject private company data into prompts. Fine-tuning alters model weights for general behavior rather than dynamically retrieving specific documents.

Correct answer: C. Interpretability is about understanding the internal mechanisms of a machine learning model, whereas explainability focuses on providing understandable reasons for the model's predictions and behaviors to stakeholders

Interpretability means a human can understand the internal mechanics driving a model's predictions. Explainability provides stakeholder-friendly justifications for outcomes, especially when the underlying model is too complex to interpret directly.

Correct answer: D. Data residency is concerned with the physical location of data storage, whereas data retention defines the policies for how long data should be stored and maintained

Data residency refers to the physical or geographical location where data is stored. Data retention dictates the lifecycle policies determining how long data is kept before eventual deletion or archiving.

Correct answer: A. Machine learning is a subset of artificial intelligence that involves training algorithms to learn from data, while artificial intelligence encompasses a wider range of technologies aimed at simulating human intelligence

Artificial intelligence is the broad discipline of simulating human intelligence, while machine learning is a specific subset focused on algorithms learning from data. A good exam cue is to remember the nesting: machine learning fits inside AI, not the other way around.

Correct answer: A. Amazon Mechanical Turk provides a marketplace for outsourcing various tasks to a distributed workforce, while Amazon Ground Truth is specifically designed for creating labeled datasets for machine learning, incorporating both automated and human labeling

Amazon Mechanical Turk is a general marketplace for outsourcing microtasks to human workers, whereas Amazon Ground Truth is a dedicated machine learning service for building datasets. Ground Truth uses automation and can route tasks to public, vendor, or private workforces.

Correct answer: A. The company should use a VPC endpoint for Amazon S3 that allows secure, private connectivity between the VPC and Amazon S3, without the need for an internet connection, ensuring data is transferred securely within the AWS network

A VPC endpoint provides private connectivity between resources inside an isolated VPC and AWS services like Amazon S3. Relying on a NAT gateway or Internet Gateway fails the requirement because those routes still push data traffic over the public internet.

Correct answer: A. SageMaker Ground Truth enables the creation of high-quality labeled datasets by incorporating human feedback in the labeling process, which can be used to improve reinforcement learning models

SageMaker Ground Truth provides human-in-the-loop capabilities to create high-quality datasets by collecting human feedback, which is essential for reinforcement learning. The other options are incorrect because Ground Truth explicitly relies on human input rather than automated synthetic data generation or real-time decision-making.

Correct answer: A. Validation sets are optional · C. Test set is used to determine how well the model generalizes

Validation sets are optional and used for hyperparameter tuning during training. Test sets are mandatory for evaluating the final model on unseen data to determine generalization. A strong distractor fails because the test set must never be used for tuning.

Correct answer: A. The developer should create a rule-based application that uses predefined mathematical rules and formulas to answer probability questions accurately

A rule-based system works best for fundamental math where outcomes rely on strict, predefined formulas. Machine learning is unnecessary here because probability calculations require exact deterministic answers rather than inferred predictions.

Correct answer: D. Amazon Inspector

Amazon Inspector automatically assesses applications for vulnerabilities and deviations from security best practices. AWS Config tracks resource configuration states, while AWS Artifact supplies compliance reports rather than automated assessments.

Correct answer: A. Medical queries chatbot · B. Customer service chatbot

Retrieval Augmented Generation powers chatbots needing factual context from private databases, like customer service or medical queries. Creative tasks like original content generation do not require factual data retrieval.

Correct answer: B. Generative AI is important because it can autonomously create novel and complex data, enhancing creativity and efficiency in various domains

Generative AI is valued because it creates novel, complex data, which boosts creativity and efficiency across diverse fields. Distractors fail because sorting data is a trivial task, and generative AI extends well beyond just gaming and entertainment.

Correct answer: C. Amazon SageMaker Data Wrangler

Amazon SageMaker Data Wrangler provides a visual interface with hundreds of built-in transformations to prepare data for machine learning without writing code. SageMaker Clarify is used for bias detection, while Feature Store simply stores and manages features.

Correct answer: C. This process is called inference, where the model uses its trained parameters to generate a prediction or output based on new input data provided by the user

Inference is the process of using a trained model to generate predictions or outputs based on new input data. Training involves adjusting model parameters using labeled data, which happens before the model is deployed.

Correct answer: B. Underfit models experience high bias, whereas, overfit models experience high variance

Underfit models suffer from high bias because they are too simple to capture the underlying trend, resulting in poor training performance. Overfit models suffer from high variance because they memorize training data but fail to generalize to new data.

Correct answer: A. Amazon Q Business guardrails support topic-specific controls to determine the web application environment's behavior when it encounters a mention of a blocked topic by an end-user · D. Amazon Q Business chat responses can be generated using model knowledge and enterprise data, or enterprise data only

Amazon Q Business allows administrators to configure topic-specific guardrails to block unwanted subjects. Additionally, administrators can control whether chat responses rely solely on enterprise data or also use the foundational model's pre-trained knowledge.

Correct answer: A. Real-time inference follows a synchronous execution mode, whereas batch inference follows an asynchronous execution mode · C. Real-time inference is used for applications requiring immediate predictions with low latency, whereas batch inference is used for processing large volumes of data at once, often with higher latency

Real-time inference operates synchronously to deliver immediate low-latency predictions for interactive applications. Batch inference operates asynchronously to process large volumes of data at once without requiring immediate responses.

Correct answer: A. Continued Pre-training · B. Fine-tuning

Amazon Bedrock supports model customization through fine-tuning using labeled data and continued pre-training using unlabeled data to adjust model weights. Retrieval Augmented Generation and prompt engineering are context-providing techniques, not weight-updating customization methods.

Correct answer: D. AWS is responsible for securing the infrastructure that runs Amazon Bedrock, while the customer is responsible for securing their data and managing access controls

Under the AWS Shared Responsibility Model, AWS secures the underlying infrastructure hosting Amazon Bedrock. Customers retain responsibility for securing their data, managing identity and access controls, and configuring application guardrails.

Correct answer: B. The company should use SageMaker Canvas, as it enables users to create machine learning models using a visual interface

Amazon SageMaker Canvas provides a no-code visual interface that allows business analysts to build, train, and deploy machine learning models without programming. Data Wrangler is strictly for data preparation, while built-in algorithms still require coding knowledge.

Correct answer: C. Average Response Time

Average response time directly measures how quickly the model processes requests and generates predictions, making it the ideal metric for runtime efficiency. Metrics like accuracy and precision evaluate predictive correctness rather than latency.

Correct answer: A. The training set is used for training the model, the validation set is used for tuning hyperparameters and model selection, and the test set is used for evaluating the final model performance

The training set teaches the model patterns, the validation set tunes hyperparameters during development, and the test set evaluates final performance on unseen data. Mixing these roles causes data leakage and results in poor generalization.

Correct answer: A. Amazon Q Developer

Amazon Q Developer accelerates software development by generating code, scanning for vulnerabilities, and upgrading applications within the integrated development environment. Distractors like Amazon Q Business focus on enterprise data retrieval rather than direct coding assistance.

Correct answer: B. Convolutional Neural Networks (CNNs)

Convolutional Neural Networks are specifically designed to process spatial data like images by automatically learning hierarchical features. Recurrent Neural Networks handle sequential data, while generative models create new content rather than classifying images.

Correct answer: A. Amazon SageMaker Feature Store

Amazon SageMaker Feature Store provides a centralized repository to store, discover, and share machine learning features securely. This avoids duplicating data pipelines, unlike Clarify which detects bias, or Data Wrangler which primarily prepares raw data.

Correct answer: D. You can use SageMaker Clarify to identify potential bias in data preparation, allowing you to detect and measure bias in datasets and models to ensure fairness and transparency in machine learning applications

Amazon SageMaker Clarify detects potential bias throughout the machine learning lifecycle, helping teams measure and mitigate bias in datasets and models. Real-time performance monitoring and drift detection belong to SageMaker Model Monitor, while tuning requires separate optimization jobs.

Correct answer: A. Specifies the minimum or maximum number of tokens to return in the generated response.

The Response length parameter dictates the minimum or maximum number of tokens the model generates. For your exam, remember that Top K and Top P control candidate selection probabilities, while Stop sequences halt generation entirely.

Correct answer: D. Gives the ability to use machine learning to generate predictions without the need to write any code

Amazon SageMaker Canvas is a no-code visual interface that lets business analysts generate ML predictions. Use this cue: JumpStart handles pre-built one-click solutions, whereas Canvas focuses on code-free model building and prediction.

Correct answer: D. NLP is used for analyzing and generating human language, such as text and speech, while Computer Vision is used for interpreting and understanding visual information from images and videos

Natural Language Processing handles human language like text and speech, whereas Computer Vision interprets visual data like images and video. A practical exam tip is mapping image recognition specifically to Computer Vision, not NLP.

Correct answer: A. Supervised machine learning involves training models with labeled data to make predictions or classify data, whereas unsupervised machine learning identifies patterns and relationships in unlabeled data

Supervised learning requires labeled data to predict outcomes, while unsupervised learning finds hidden patterns in unlabeled data. Remember the core exam cue: if the data has a specific target or label, it strictly indicates supervised learning.

Correct answer: B. RLHF is a technique used to train AI models using human feedback to refine their behavior, whereas A2I is an AWS service that provides a human review of machine learning predictions to improve model accuracy and reliability

Reinforcement Learning from Human Feedback trains models using human input to refine behavior, whereas Amazon A2I provides ongoing human review of ML predictions. Remember A2I specifically handles looping human audits into production workflows.

Correct answer: B. Diffusion models create new data by iteratively making controlled random changes to an initial data sample

Diffusion models generate new data by adding and then iteratively removing controlled random noise from a sample. For your exam, associate competing neural networks with GANs, and latent space compression specifically with variational autoencoders.

Correct answer: D. Structured data is organized in a predefined manner, often in rows and columns, making it easy to search and analyze, while unstructured data lacks a specific format and includes data like text, images, and videos

Structured data is organized in predefined formats like rows and columns, whereas unstructured data lacks specific formatting and includes raw text or media. A practical exam cue is mapping tabular spreadsheet data to structured categorization.

Correct answer: B. Hallucination

Hallucination occurs when a generative AI model produces convincing but factually incorrect information. For the exam, distinguish hallucination from explainability or fairness, which respectively deal with understanding model decisions and mitigating bias.

Correct answer: C. Overfitting occurs when the model is overly complex and captures noise or random fluctuations in the training data rather than the underlying patterns

Overfitting happens when an overly complex model memorizes noise and random fluctuations in the training data instead of learning general patterns. Using fewer features indicates a simpler model, which is typically associated with underfitting rather than overfitting.

Correct answer: A. Reinforcement learning is used for optimizing complex systems such as robotics, game playing, and industrial automation by learning optimal actions through trial and error

Reinforcement learning optimizes complex systems by using an agent that learns optimal actions through trial and error. Distinguish this from unsupervised learning, which clusters unlabeled data, and supervised learning, which predicts outcomes using historical trends.

Correct answer: B. Improving model interpretability and transparency may sometimes involve trade-offs with model performance, as simpler models are often easier to interpret but may not achieve the highest performance

Increasing model interpretability often involves trade-offs with performance because simpler models are easier to explain but might miss complex patterns. Complex deep learning models generally offer higher accuracy but act as opaque black boxes that reduce transparency.

Correct answer: C. The company's data is not used to improve the base Foundation Models (FMs) and it is not shared with any model providers

Amazon Bedrock does not use customer prompts or outputs to improve the base foundation models, nor does it share your data with third-party model providers. This data isolation ensures compliance for strict workloads like healthcare.

Correct answer: A. Human model evaluation is valuable for assessing qualitative aspects of the model, whereas, automatic model valuation is valuable for assessing quantitative aspects of the model · E. Automatic model evaluation provides model scores that are calculated using various statistical methods such as BERT Score and F1

Automatic evaluation assesses quantitative metrics using statistical scoring methods like BERT Score and F1, while human evaluation handles qualitative aspects. Option D fails because human workforces require custom prompt datasets, not built-in ones.

Correct answer: C. FM is an AI model with a large number of parameters and trained on a massive amount of diverse data, whereas, model customization is the process of using training data to adjust the model parameter values in a base model to create a custom model

A foundation model is pre-trained on diverse, massive datasets for broad use cases, whereas customization adjusts the model parameters using your specific training data. Options reversing these definitions fail to capture the core distinction.

Correct answer: D. An ML algorithm is a set of mathematical instructions for solving a specific type of problem, while an ML model is the output of the algorithm after being trained on data

An algorithm represents the mathematical instructions or procedures, whereas the model is the specific output generated after training on data. Distractors fail because they confuse mathematical procedures with data storage or network security functions.

Correct answer: D. Machine learning involves training algorithms on large datasets to identify patterns and make predictions or decisions based on new data

Machine learning involves training algorithms on large datasets so they can identify patterns and make predictions on new data. Distractors fail because they incorrectly claim the process uses predefined rules, requires no data, or relies strictly on hardware.

Correct answer: D. Influences the number of most-likely candidates that the model considers for the next token

The Top K parameter limits the token selection pool to the most likely candidates, directly controlling text diversity. Do not confuse this with Top P, which uses a probability percentage cutoff, or Temperature, which scales the overall creativity.

Correct answer: B. In traditional machine learning, a data scientist manually determines the set of relevant features that the software must analyze, whereas in deep learning, the data scientist gives only raw data to the software and the deep learning network derives the features by itself · D. Deep learning is a subset of machine learning that uses neural networks with many layers to learn from large amounts of data, while traditional machine learning algorithms often require feature extraction and can use various methods such as decision trees or support vector machines

Deep learning automatically extracts features from raw data using neural networks, whereas traditional machine learning requires manual feature engineering. Remember that deep learning typically demands large datasets and computational power, unlike simpler traditional models.

Correct answer: D. LLMs are used for generating human-like text, translating languages, summarizing text, and answering questions based on large datasets

Large language models excel at text-based tasks like summarization, translation, and answering questions. They are not designed to generate multimedia assets like 3D models, video, or audio directly, which require specialized generative architectures.

Correct answer: B. K-Means is an unsupervised learning algorithm used for clustering data points into groups, while KNN is a supervised learning algorithm used for classifying data points based on their proximity to labeled examples

K-Means is an unsupervised clustering algorithm that groups unlabelled data, whereas K-Nearest Neighbors is a supervised classification technique relying on labeled examples. Avoid confusing these two solely because they both start with the letter K.

Correct answer: B. RAG refers to querying and retrieving information from a data source to augment a generated response to a prompt, whereas, Agent refers to an application that carries out orchestrations through cyclically interpreting inputs and producing outputs by using a foundation model

Retrieval Augmented Generation strictly retrieves external data to enhance a prompt, while an agent actively orchestrates multistep tasks by interpreting inputs and executing actions. Agents often utilize data retrieval as just one tool within a broader automated workflow.

Correct answer: A. By learning patterns from existing data and using algorithms to generate new content that mimics those patterns

Generative AI creates new content by learning underlying patterns from training data and using those learned distributions to generate novel, similar outputs. For the exam, remember that generative AI goes beyond rigid templates or traditional programming; it actively learns from data rather than relying on hard-coded rules.

Correct answer: C. Data residency refers to where data is physically stored, while data logging tracks data access and changes over time

Data residency dictates the physical or geographical location where data is stored, ensuring compliance with local regulations. Data logging records access and modifications to that data, creating a necessary audit trail for security and troubleshooting.

Correct answer: C. The company should use benchmark datasets, which are pre-compiled, standardized datasets specifically designed to test for biases and discrimination in model outputs

Standardized benchmark datasets provide a reliable, pre-compiled resource to test large language models for bias without requiring extensive manual setup. Generating synthetic data or relying on manual human review demands significantly more administrative effort and resources to maintain.

Correct answer: C. Feature extraction reduces the number of features by transforming data into a new space, while feature selection reduces the number of features by selecting the most relevant ones from the existing features

Feature extraction transforms original data into a completely new set of features, such as using principal component analysis to reduce dimensionality. Feature selection simply chooses the most relevant variables from the existing dataset without altering their underlying structure.

Correct answer: D. AWS Config

AWS Config continuously monitors and records resource configurations, allowing automated compliance checks against your desired baselines. While AWS Audit Manager and Artifact handle compliance reporting and audits, they lack the active, continuous resource configuration tracking required here.

Correct answer: C. AWS Lambda · E. Amazon Rekognition

AWS Lambda and Amazon Rekognition are regional services, meaning you provision and manage them within a specific AWS region. Distractors like IAM, CloudFront, and Web Application Firewall operate globally to provide consistent security and edge delivery across all regions.

Correct answer: B. Prompt engineering, Retrieval Augmented Generation (RAG), Fine-tuning

Prompt engineering is the simplest customization method, followed by Retrieval Augmented Generation which requires vector stores, while fine-tuning demands deep data science expertise. Fine-tuning alters model weights, making it the most complex and resource-intensive technique of the three.

Correct answer: A. Guardrails for Amazon Bedrock

Guardrails for Amazon Bedrock is the correct choice because it lets you block harmful content and mask PII directly within your generative AI workflows. Knowledge Bases is for Retrieval Augmented Generation, not content filtering or privacy protection.

Correct answer: C. To offer transparency and information about the intended use, limitations, and potential impacts of AWS AI services, helping users implement Responsible AI practices

AWS AI service cards provide transparency by detailing intended uses and limitations to support Responsible AI practices. Distractors about technical documentation or a prepaid marketplace fail because the cards strictly focus on ethical implementation guidance.

Correct answer: B. Yes, it is possible to increase both bias and variance, but this typically leads to a model that performs poorly due to both underfitting and overfitting

Bias and variance can both increase simultaneously, which usually yields a poorly performing model that underfits and overfits. While they often exhibit a tradeoff, removing training data or using a flawed algorithm can incorrectly increase both errors.

Correct answer: B. Threat detection involves real-time monitoring and identification of active threats, whereas vulnerability management is about identifying, assessing, and mitigating security weaknesses

Threat detection focuses on identifying active malicious activity in real time, whereas vulnerability management proactively identifies and mitigates security weaknesses. Associate GuardDuty with active threats and Inspector with vulnerability management to eliminate reversed distractors.

Correct answer: B. Requests with large payload sizes up to 1GB and long processing times

Amazon SageMaker Asynchronous Inference specifically handles large payloads up to one gigabyte and long processing times by queueing requests. Remember that batch transform fits offline dataset predictions, while serverless inference suits intermittent traffic.

Correct answer: D. The company should use AWS Artifact to facilitate on-demand access to AWS compliance reports and agreements, as well as allow users to receive notifications when new compliance documents or reports, including ISV compliance reports, are available

AWS Artifact provides on-demand access to security and compliance reports, including third-party Independent Software Vendor reports. You can subscribe to notifications within Artifact, whereas Audit Manager and Config evaluate internal resource configurations.

Correct answer: D. Precision, Recall and F1-Score

Precision, recall, and the F1 score are the standard metrics for evaluating classification model performance. Mean absolute error and root mean squared error apply to regression problems, while latency and throughput measure infrastructure performance.

Correct answer: C. Amazon Bedrock

Amazon Bedrock is the underlying managed service that powers the generative AI capabilities of Amazon Q Business. Remember that Amazon Kendra provides the enterprise search functionality, while Q Apps allows users to build custom AI applications.

Correct answer: B. A data scientist selects features for a machine learning model based on their personal beliefs about which attributes are important, leading to a biased model

A data scientist incorporating personal beliefs into feature selection directly illustrates human bias. Option D describes algorithmic bias because the skew originates from historical data trends rather than the developer's subjective choices.

Correct answer: B. The company should use batch inference, thereby allowing it to run multiple inference requests in a single batch

Batch inference processes large volumes of data asynchronously, making it the most cost-effective choice when immediate responses are not required. On-demand inference costs more for frequent workloads since it lacks bulk processing discounts.

Correct answer: A. Implementing guardrails and enhancing transparency for generative AI applications

Implementing guardrails and maintaining transparency are core best practices to ensure responsible generative AI deployment and user safety. Ignoring continuous monitoring or prioritizing speed over ethics introduces significant operational and compliance risks.

Correct answer: B. Clustering · E. Dimensionality reduction

Clustering and dimensionality reduction are classic unsupervised learning techniques used to find patterns or reduce features in unlabeled data. Decision trees represent supervised learning, while neural networks can operate across multiple learning paradigms.

Correct answer: B. The company should use Amazon QuickSight, a business intelligence (BI) service that allows users to easily create and share interactive dashboards and visualizations from various data sources, including up-to-date sales data, enabling real-time insights and reporting

Amazon QuickSight is a dedicated business intelligence service for creating interactive dashboards and visualizing business data like sales metrics. CloudWatch monitors infrastructure, while SageMaker tools focus strictly on machine learning data preparation and modeling.

Correct answer: A. The company should create a service role for Amazon Bedrock for each team, granting access only to the specific team's clients data in Amazon S3

Creating distinct AWS service roles for each team ensures Amazon Bedrock only accesses the specific client data authorized for that team. Relying on broad identity policies or post-incident logging fails to proactively isolate data at the service level during model inference.

Correct answer: A. Risk management

Risk management is the correct answer because it involves proactively identifying threats and applying mitigations to secure generative AI solutions. Governance handles policies and compliance reporting, which differs from the active threat modeling covered by risk management.

Correct answer: A. Data collection, Data preprocessing, Model training, Model evaluation

The standard machine learning pipeline begins with data collection and preprocessing to prepare the dataset, followed by model training and final evaluation. Any sequence placing evaluation or training before data preparation is logically invalid.

Correct answer: D. Supervised learning, Unsupervised learning, Deep Learning

AWS categorizes the three main machine learning types as supervised learning, unsupervised learning, and deep learning for certification purposes. Although reinforcement learning is a valid ML paradigm, stick with this specific taxonomy for the exam.

Correct answer: C. While CNNs are used for single image analysis, RNNs are used for video analysis

Convolutional neural networks excel at spatial data processing, making them ideal for analyzing single images. Recurrent neural networks process sequential information, making them better suited for video analysis or time-series data. Distractors incorrectly swap these core architectural strengths.

Correct answer: A. Building and training a generative AI model from scratch

Building a model from scratch places you in Scope 5 of the Generative AI Security Scoping Matrix, requiring full ownership of infrastructure and security. Consuming a public service shifts most security responsibilities to the provider, minimizing your operational burden.

Correct answer: A. Amazon Comprehend

Amazon Comprehend is the correct service because its natural language processing capabilities can automatically detect and redact personally identifiable information in text. Amazon Kendra is an enterprise search service, so it requires clean data rather than serving as the redaction tool itself.

Correct answer: A. Labeled data is annotated with output labels that provide specific information about each data point and is used for supervised learning, whereas, unlabeled data lacks such annotations and is used for unsupervised learning

Labeled data includes specific annotations or tags used to train supervised learning models for predicting known outcomes. Unlabeled data lacks these tags and is used for unsupervised learning tasks, where the model independently finds hidden patterns or structures within the dataset.

Correct answer: A. Amazon Comprehend Medical

Amazon Comprehend Medical uses natural language processing to extract medical conditions, medications, and protected health information from unstructured clinical text. Standard Amazon Comprehend is incorrect because it is built for general text analysis rather than specialized healthcare data.

Correct answer: A. Leverage Amazon Bedrock to make a separate copy of the base FM model and train this private copy of the model using the labeled training dataset

Amazon Bedrock fine-tuning creates a private, separate copy of the base foundation model trained securely on your labeled data. You cannot alter the original base model for all users, nor can you use Bedrock to train a new model entirely from scratch.

Correct answer: B. Amazon Q Developer can suggest code snippets, providing developers with recommendations for code based on specific tasks or requirements

Amazon Q Developer accelerates coding by providing real-time code snippets and recommendations directly within your integrated development environment. It is an assistant for writing code, not a deployment pipeline tool or a service for building custom SageMaker machine learning models.

Correct answer: B. The company should use a Generative Adversarial Network (GAN) for creating realistic synthetic data while preserving the statistical properties of the original data

Generative Adversarial Networks, or GANs, excel at creating realistic synthetic data while preserving the statistical properties of the original records. Support Vector Machines and Convolutional Neural Networks are primarily used for classification and feature extraction rather than generation.

Correct answer: C. Amazon SageMaker Model Cards

Amazon SageMaker Model Cards consolidate critical model details like intended use cases, training metrics, and evaluation observations into a single report. SageMaker Clarify identifies bias, whereas Model Monitor tracks ongoing production performance.

Correct answer: B. Amazon Textract

Amazon Textract uses machine learning to automatically extract text, handwriting, and structured data from scanned documents. For the exam, match Rekognition with image and video analysis, and match Transcribe with converting audio speech into text.

Correct answer: C. Guardrails helps control the interaction between users and FMs by filtering undesirable and harmful content, whereas, watermark detection identifies if an image was created by the Amazon Titan Image Generator model on Bedrock

Guardrails evaluates and filters harmful user inputs and model outputs within Amazon Bedrock. Watermark detection serves a separate purpose by verifying if an image was generated by the Amazon Titan Image Generator, rather than actively filtering conversational prompts.

Correct answer: C. Amazon Q Business

Amazon Q Business securely connects to enterprise data to answer questions, summarize reports, and generate content. Amazon Q Developer targets coding tasks, whereas Amazon Q in Connect is designed specifically for customer service agent assistance.

Correct answer: C. Amazon Forecast

Amazon Forecast uses machine learning to deliver highly accurate time-series forecasts for metrics like foot traffic. Amazon Personalize handles real-time product recommendations, while Amazon Lex builds conversational chatbots rather than predicting future numerical demand.

Correct answer: B. Amazon Augmented AI (A2I)

Amazon Augmented AI provides built-in human review workflows to easily audit machine learning predictions and route low-confidence cases to reviewers. SageMaker Ground Truth is primarily used for initial data labeling rather than ongoing prediction validation.

Correct answer: D. Amazon Comprehend

Amazon Comprehend uses natural language processing to extract key insights, entities, and phrases from unstructured text like legal documents. Amazon Transcribe converts speech to text, while Amazon Rekognition analyzes images rather than textual data.

Correct answer: A. Amazon SageMaker Ground Truth

Amazon SageMaker Ground Truth provides comprehensive human-in-the-loop capabilities by allowing human annotators to review, label, and provide feedback on machine learning data. SageMaker Clarify focuses on bias detection, not the active integration of human feedback during model development.

Correct answer: D. Real-time hosting services

Real-time hosting services provide persistent endpoints designed for low-latency, interactive predictions required during synchronous patient check-ins. Batch transform handles entire datasets offline, while asynchronous inference is intended for large payloads or long processing times.

Correct answer: D. It ensures data privacy and compliance by tracking the flow and transformation of data

Maintaining data lineage tracks the flow and transformation of data to ensure privacy, security, and regulatory compliance for highly sensitive industries like finance. It does not directly reduce storage costs or optimize machine learning model performance, though it improves auditability.

Correct answer: D. Multi-class classification assigns each instance to one of several possible classes, while multi-label classification assigns each instance to one or more classes

Multi-class classification restricts each instance to exactly one category, while multi-label classification allows instances to belong to multiple categories simultaneously. Media content often requires multi-label classification because a single video might be categorized as both sports and technology.

Correct answer: B. Amazon Transcribe medical

Amazon Transcribe Medical is specifically trained to understand medical terminologies and clinical conversations, making it ideal for healthcare compliance. Standard Amazon Transcribe handles general speech-to-text tasks but lacks the specialized medical vocabulary required for accurate clinical documentation.

Correct answer: A. Amazon Polly

Amazon Polly is a text-to-speech service that uses deep learning to synthesize natural-sounding human speech in multiple languages. Amazon Translate converts text between languages but does not generate audio, while Amazon Lex builds conversational chatbots.

Correct answer: C. Generative models focus on generating new data from learned patterns, whereas discriminative models classify data by distinguishing between different classes

Generative models learn underlying data distributions to create entirely new content, whereas discriminative models focus on finding decision boundaries to classify existing data. Discriminative models are strictly used for tasks like classification or prediction, not generating new data.

Correct answer: D. Amazon Rekognition

Amazon Rekognition provides pretrained computer vision APIs to detect objects and text in images or video. Textract is optimized for OCR on structured documents rather than reading text from real-world scenes like license plates.

Correct answer: B. Amazon Bedrock

Amazon Bedrock offers managed access to leading foundation models and supports private customization using your proprietary data. AWS Inferentia is hardware for inference acceleration, whereas Amazon Q applications are purpose-built assistants without fine-tuning capabilities.

Correct answer: D. Amazon SageMaker Clarify

Amazon SageMaker Clarify provides feature importance graphs to explain how specific inputs influence model predictions during development and inference. Canvas is a no-code tool, JumpStart deploys models, and Ground Truth handles data labeling.

Correct answer: G. A-2, B-3, C-1

Textract provides OCR for documents, Forecast predicts business metrics, and Kendra delivers ML-powered enterprise search. Matching these core capabilities correctly helps differentiate purpose-built AWS AI services quickly.

Correct answer: C. Amazon Textract

Amazon Textract is optimized to extract printed text, handwriting, and structured data from scanned documents like invoices. Rekognition focuses on identifying sparse text in real-world images rather than processing dense structured forms.

Correct answer: D. Amazon SageMaker Model Dashboard

The SageMaker Model Dashboard provides a centralized view by aggregating metadata, endpoint status, and monitoring alerts. Feature Store manages ML inputs, Data Wrangler prepares data, and JumpStart deploys foundation models.

Correct answer: D. The company should use Accuracy, which measures the proportion of correctly predicted instances (both true positives and true negatives) out of the total number of instances

Accuracy measures the overall proportion of correct binary predictions, fitting the requirement to see correct outcomes. RMSE and R-squared are regression metrics, while F1 Score is typically reserved for evaluating imbalanced datasets.

Correct answer: D. Stable Diffusion

Stable Diffusion is the correct choice because it is a foundational image generation model available in Amazon Bedrock that excels at creating photorealistic images from text prompts. Claude, Llama, and Jurassic are large language models built primarily for text generation, reasoning, and summarization tasks.

Correct answer: B. Amazon Bedrock provides foundational models for generative AI applications, whereas Amazon SageMaker JumpStart offers pre-built solutions and one-click deployment for various machine learning models

Amazon Bedrock is a fully managed service that provides access to foundational models from leading AI companies via an API to build generative AI applications. Amazon SageMaker JumpStart provides pre-built solutions and one-click deployment for a broader range of traditional machine learning models.

Correct answer: A. AWS Trainium instances

AWS Trainium instances are explicitly designed to offer the highest energy efficiency and performance per watt for deep learning training workloads. Standard accelerated computing instances like P or G types rely on traditional GPUs, which consume significantly more power and lack this specific optimization.

Correct answer: B. Implement robust auditing processes for AI outputs · C. Establish ethical AI guidelines for developers to follow

Establishing ethical guidelines and implementing robust auditing processes provide the necessary proactive framework for responsible AI development and accountability. The distractors suggest risky practices, like relying solely on user feedback or unrestricted fine-tuning, which fail to ensure proper governance.

Correct answer: A. Leverage AWS Trainium for high-performance, cost-effective Deep Learning training. Leverage AWS Inferentia for the deep learning (DL) and generative AI inference applications

AWS Trainium is purpose-built for high-performance and cost-effective deep learning model training. Conversely, AWS Inferentia is specifically designed to provide the lowest cost and highest performance for deep learning and generative AI inference workloads on Amazon EC2.

Correct answer: A. Amazon Connect

Amazon Connect provides an easy-to-use, cloud-based contact center that can be set up in minutes and supports agents anywhere. While Amazon Lex builds conversational chatbots, it is not the contact center platform itself.

Correct answer: C. Zero shot Prompting

Zero-shot prompting asks a model to perform a task without providing specific examples in the prompt. Few-shot prompting is incorrect because it relies on giving the model multiple examples to guide its output.

Correct answer: A. Amazon Q in QuickSight

Amazon Q in QuickSight is a generative business intelligence assistant that lets business analysts build dashboards using natural language. Amazon Q Business targets enterprise data search, while Q Developer focuses on coding tasks.

Correct answer: D. The Large Language Models (LLMs) are non-deterministic

Large Language Models are non-deterministic, meaning the same prompt can yield different outputs. Temperature controls this variability; lower values make responses more deterministic, while higher values increase creativity.

Correct answer: B. Amazon SageMaker Data Wrangler

Amazon SageMaker Data Wrangler provides a visual interface to prepare data for machine learning, including creating train, validation, and test splits. Ground Truth handles data labeling, whereas Clarify detects bias.

Correct answer: B. Amazon SageMaker Data Wrangler

SageMaker Data Wrangler reduces data preparation time by letting you import, clean, and transform tabular data from multiple sources visually. Feature Store only manages ML features, while Ground Truth is strictly for data labeling.

Correct answer: C. Serverless Inference

Serverless Inference is ideal for intermittent traffic because it scales to zero and tolerates cold starts, eliminating infrastructure management. Real-time endpoints keep persistent instances running, which costs more during idle periods.

Correct answer: A. Implement dynamic prompt engineering to customize responses based on user characteristics like age

Dynamic prompt engineering lets you alter input instructions on the fly to generate age-appropriate responses without modifying the model. Fine-tuning or retraining is unnecessary and inefficient for simple stylistic adjustments.

Correct answer: B. The company should use Amazon S3, which is a scalable object storage service fully integrated with Amazon Bedrock

Amazon S3 is the primary storage service for Amazon Bedrock model customization, providing scalable and fully integrated dataset access. EBS and EFS are block and file storage services, which are not natively used for Bedrock training data.

Correct answer: D. Amazon SageMaker Model Cards

Amazon SageMaker Model Cards are used to document critical model details like intended use and risk rating, supporting governance and reporting. Avoid Amazon SageMaker Model Monitor here, as its primary function is tracking production model performance drift rather than establishing usage guidelines.

Correct answer: A. SageMaker Feature Store

SageMaker Feature Store provides a centralized repository to store, share, and manage machine learning features across teams. Model Monitor handles production drift, Data Wrangler prepares raw data, and Clarify detects bias in datasets and models.

Correct answer: D. Create a prompt template that teaches the LLM to detect attack patterns

Using carefully designed prompt templates helps the model recognize malicious patterns and reject inappropriate instructions. Restricting prompt length or disabling inputs fails to address the core issue of adversarial prompt engineering without breaking functionality.

Correct answer: B. Use a benchmark dataset for evaluation

Benchmark datasets provide a standardized, objective baseline to rigorously evaluate model accuracy against broader industry standards. Testing on training data causes overfitting, while deploying untested models into production or relying on random manual checks introduces unnecessary risk.

Correct answer: C. The company should use object detection, which involves identifying and locating specific objects within an image

Object detection is the correct technique because it identifies and locates specific objects within images, which fits the animal tracking use case. Face recognition only handles human faces, whereas thermal imaging shows heat rather than visual features needed for precise classification.

Correct answer: A. Generate recommendations for items that are similar to an item you specify

Amazon Personalize is the correct choice because it generates item recommendations based on user behavior and metadata. Amazon Kendra handles enterprise search, while Amazon Textract extracts layout elements from documents, making them incorrect for a recommendation use case.

Correct answer: A. Use GroundTruth Plus to label the data

Amazon SageMaker Ground Truth Plus provides an expert managed workforce and quality assurance workflows to ensure high annotation accuracy. Relying solely on pre-trained models or small internal teams risks introducing inconsistencies and biases into the training dataset.

Correct answer: A. Reinforcement Learning

AWS DeepRacer uses reinforcement learning to train models through trial and error, rewarding correct actions within a simulated environment. Deep learning is a broader machine learning category, whereas reinforcement learning specifically defines the DeepRacer training methodology.

Correct answer: D. Amazon SageMaker Role Manager, Amazon SageMaker Model Cards, Amazon SageMaker Model Dashboard

Role Manager controls access, Model Cards document model details for transparency, and Model Dashboard tracks production behavior. Clarify evaluates bias and explainability, while Model Monitor detects drift, so neither fits the strict governance documentation need.

Correct answer: B. Amazon Textract · D. Amazon Rekognition

Amazon Textract and Amazon Rekognition both provide optical character recognition capabilities. Textract is optimized for dense documents, while Rekognition handles sparse text within real-world images. Polly, Lex, and Comprehend process or generate text but cannot extract it from images.

Correct answer: A. Bidirectional Encoder Representations from Transformers (BERT) based Model

A BERT-based transformer model uses bidirectional context to accurately predict missing words by analyzing surrounding text. Rule-based models lack contextual awareness, while clustering models group data rather than generating text predictions.

Correct answer: C. Influences the likelihood of the model selecting lower-probability outputs, thereby impacting the creativity of the model's output

The temperature parameter controls the randomness or creativity of a model's responses by influencing the likelihood of selecting lower-probability outputs. For the exam, remember that Top K limits the number of candidates considered, while Top P limits the cumulative percentage of candidates.

Correct answer: A. Average Call Duration

Average Call Duration is the correct metric because the stated goal is to improve efficiency and reduce the time agents spend on calls. While First-Call Resolution Rate is an important call center metric, it measures overall customer success rather than the direct time savings targeted by the chatbot.

Correct answer: B. The AWS DeepRacer vehicle is a Wi-Fi enabled, physical vehicle that can drive itself on a physical track

AWS DeepRacer is a physical, Wi-Fi-enabled vehicle that uses reinforcement learning to drive autonomously on physical tracks. It relies on reinforcement learning algorithms, not supervised learning.

Correct answer: A. Amazon Textract · E. Amazon Comprehend · F. Generative AI powered summarization chatbot

Amazon Textract extracts raw text from scanned documents, Amazon Comprehend analyzes that text to find entities, and generative AI summarizes the findings. Watch the wording: a generic chatbot is vague, but combining Textract and Comprehend is the definitive AWS approach for intelligent document processing.

Correct answer: C. Describes how a model should be used in a production environment

Amazon SageMaker Model Cards document essential model details such as intended uses, risk ratings, and evaluation metrics for governance. The wording is slightly ambiguous because option B is technically true, but option C captures the primary purpose of guiding production usage.

Correct answer: C. Fix bias by balancing the dataset

Amazon SageMaker Data Wrangler includes built-in transforms to balance datasets, helping address data skew and mitigate potential bias during data preparation. For the exam, remember that Feature Store shares variables and Model Monitor tracks production drift.

Correct answer: A. All

Amazon SageMaker Studio supports multiple integrated development environments, including JupyterLab, Code Editor, and RStudio. Choosing individual options fails because the platform provides this comprehensive suite of tools to accommodate diverse data science workflows.

Correct answer: B. The company should use GPT (Generative Pre-trained Transformer), to interpret natural language inputs and generating coherent outputs, such as SQL queries, by leveraging its understanding of language patterns and structures

Generative Pre-trained Transformers are specifically designed to process natural language and generate human-like text, making them ideal for translating plain English into SQL queries. Amazon Comprehend is built for extracting insights from existing text, not generating new SQL statements.

Correct answer: D. None

Amazon SageMaker Automatic Model Tuning can automatically choose hyperparameter ranges, search strategy, and the number of jobs based on your objective metric. For the exam, remember that while you can manually configure these parameters, none are strictly mandatory to initiate tuning.

Correct answer: D. Amazon Q Developer can be used in integrated development environments (IDEs) as well as the AWS Management Console

Amazon Q Developer operates directly within integrated development environments and across the AWS Management Console to assist with coding and AWS queries. Do not assume it is restricted to just the IDE; it also provides architectural guidance directly inside the console dashboard.

Correct answer: B. The company should instruct the model to stick to the prompt by adding explicit instructions to ignore any unrelated or potentially malicious content

Adding explicit instructions within the prompt is a straightforward prompt engineering technique to help guide model behavior and reject malicious inputs. However, for robust production workloads on Amazon Bedrock, implementing automated guardrails provides stronger content filtering than prompt instructions alone.