AWS Certified AI Practitioner (AIF-C01) — All Questions

Which statement best captures the relationship between artificial intelligence, machine learning, and deep learning?

• a.Artificial intelligence is a narrow branch that sits entirely inside the field of machine learning
• b.Deep learning is the broad umbrella term that fully contains machine learning and artificial intelligence
• c.Machine learning and deep learning are unrelated fields that each compete against artificial intelligence
• d.Deep learning is a subset of machine learning, which is itself a subset of artificial intelligence✓

AI is the broad goal of machines performing tasks that seem intelligent. Machine learning is a subset of AI in which systems learn patterns from data. Deep learning is a further subset of ML that uses multi-layer neural networks.

A team labels thousands of emails as 'spam' or 'not spam' and trains a model to predict the label for new emails. Which learning paradigm is this?

• a.Reinforcement learning, because the model earns a reward signal for each email that it reads
• b.Supervised learning, because the model learns from examples that include the correct output labels✓
• c.Transfer learning, because the model simply reuses stored weights taken from a completely unrelated source task
• d.Unsupervised learning, because the model groups the incoming emails without guidance

Supervised learning trains on labeled examples (input plus the known correct output) so the model can predict labels for unseen inputs. Spam classification with labeled emails is a classic supervised task.

A retailer wants to segment customers into natural groups without any predefined categories. Which approach fits best?

• a.Supervised learning using a regression algorithm to predict one continuous target value
• b.Reinforcement learning using a reward function that scores each sequential action taken
• c.Unsupervised learning using a clustering algorithm to group similar customers together✓
• d.A rule-based expert system using fixed hand-written thresholds

Unsupervised learning finds structure in unlabeled data. Clustering groups similar records together without predefined labels, which matches customer segmentation where the groups are not known in advance.

In machine learning, what is the difference between a feature and a label?

• a.A feature is the model's final prediction; a label is the raw input data that is fed to the model
• b.A feature is an input variable used for prediction; a label is the target output being predicted✓
• c.A feature is used only in deep learning, while a label is used only in classical machine learning models
• d.A feature and a label are simply two interchangeable names for exactly the same input data column

Features are the input variables (columns) the model reads. The label is the target value the model learns to predict. In supervised training every example pairs features with a known label.

A model performs almost perfectly on its training data but poorly on new, unseen data. What is this called?

• a.Convergence, where the model has finished training and reached its stable optimal parameter state
• b.Underfitting, where the model stays far too simple to capture even the training data itself
• c.Overfitting, where the model memorizes training noise instead of learning the general pattern✓
• d.Regularization, where the model is penalized for excess complexity

Overfitting occurs when a model learns the training data too closely, including its noise, and fails to generalize. A large train-versus-test performance gap is the classic symptom. Underfitting is the opposite problem.

Which task is an example of a regression problem rather than a classification problem?

• a.Deciding whether an incoming card transaction is fraudulent or legitimate
• b.Sorting written product reviews into positive, neutral, or negative sentiment buckets
• c.Predicting the future selling price of a house as a continuous dollar amount✓
• d.Identifying which one of ten animal species is in a photo

Regression predicts a continuous numeric value, such as a house price. Classification assigns inputs to discrete categories. The other options each choose among fixed classes and are therefore classification tasks.

Why do practitioners split a dataset into separate training and test sets?

• a.To guarantee, with certainty, that the resulting model can never overfit the data at all
• b.To completely remove any need for cleaning, formatting, or otherwise preparing the raw data
• c.To measure how well the model generalizes to data it did not see while training✓
• d.To make each training run finish faster by using far fewer total records overall

Holding out a test set lets you estimate real-world performance on unseen data. Evaluating on data used for training would give an overly optimistic score and hide overfitting.

Which AWS service is a fully managed platform for building, training, and deploying custom machine learning models at scale?

• a.Amazon SageMaker AI, the managed platform for the end-to-end model lifecycle✓
• b.Amazon CloudWatch, a service for monitoring metrics, logs, and operational alarms
• c.Amazon Rekognition, a service for detecting objects and faces inside images
• d.Amazon Route 53, a scalable and highly available Domain Name System web service

Amazon SageMaker AI is the managed platform for the end-to-end ML lifecycle: data prep, training, tuning, and deployment. Rekognition is a pre-built vision service, while CloudWatch and Route 53 are not ML platforms.

An artificial neural network is loosely inspired by which biological structure?

• a.The twisted double-helix structure of the DNA molecules found inside every cell
• b.The slowly settling layered sediment that forms many common geological rock beds
• c.The spreading branch-and-root systems that large trees rely on to gather up their water
• d.Interconnected neurons in the brain that pass signals across weighted connections✓

Artificial neural networks are modeled on biological neurons: simple units connected by weighted links that pass and transform signals across layers. Depth (many layers) is what makes a network 'deep'.

When is a traditional rule-based system often a better choice than a machine learning model?

• a.When a very large labeled dataset exists but almost no domain expertise is available
• b.When the logic is well understood and can be written as clear deterministic rules✓
• c.When the problem is recognizing thousands of different objects across millions of images
• d.When the underlying patterns are complex and constantly changing

If a problem follows clear, stable rules that experts can write down, a deterministic rule-based system is simpler, cheaper, and more transparent than ML. ML shines when patterns are complex or hard to specify by hand.

What is the primary purpose of the inference phase in a machine learning workflow?

• a.Collecting and carefully labeling all of the raw data before any model yet exists
• b.Using an already-trained model to generate predictions on brand-new input data✓
• c.Adjusting the model's internal weights repeatedly in order to reduce the training error
• d.Splitting the full dataset into separate training, validation, and holdout test portions

Inference is when a trained model is put to work generating predictions on new inputs. Adjusting weights happens during training, not inference. Data collection and splitting are earlier preparation steps.

What most clearly distinguishes generative AI from traditional discriminative machine learning?

• a.Generative AI is always faster and cheaper to run than every other type of machine learning model
• b.Generative AI creates new content such as text or images, not just labels for existing data✓
• c.Generative AI needs no training data whatsoever and can function from nothing but a short prompt
• d.Generative AI is able only to sort each of its inputs into a set of predefined fixed categories

Generative models produce new artifacts (text, images, audio, code) that resemble their training data. Discriminative models instead label or score existing inputs. Generative AI still needs training data and is often compute-intensive.

In the context of large language models, what is a token?

• a.A numeric confidence score that rates how truthful each given answer is likely
• b.A complete source document stored in the model's memory
• c.A chunk of text, such as a word or sub-word piece, processed as one unit✓
• d.A security credential that authorizes an API call

LLMs break text into tokens (words or sub-word fragments) and process them as units. Token counts drive context limits and pricing. This is unrelated to auth tokens, which are security credentials.

What is a foundation model?

• a.A specialized database index that speeds up structured queries
• b.A rule-based engine that only stores hand-written if-then statements from experts
• c.A small model trained entirely from scratch for one single narrowly defined task
• d.A large model pre-trained on broad data that adapts to many downstream tasks✓

Foundation models are large models pre-trained on massive, broad datasets, giving them general capabilities that can be adapted (via prompting or fine-tuning) to many tasks, rather than being built for a single narrow purpose.

Which AWS service provides access to foundation models from multiple providers through a single serverless API?

• a.Amazon Comprehend, a natural language service that finds named entities in text
• b.Amazon Polly, a service that turns text into speech
• c.Amazon Textract, a service that extracts text from files
• d.Amazon Bedrock, a unified serverless gateway to many foundation models✓

Amazon Bedrock offers a unified serverless API to foundation models from providers such as Anthropic, Meta, Cohere, and Amazon. Comprehend, Textract, and Polly are single-purpose AI services, not FM access layers.

What does it mean when a large language model 'hallucinates'?

• a.It slows down noticeably because the input prompt it received was simply far too long
• b.It stores the user's private personal data permanently without ever obtaining any consent
• c.It produces fluent, confident output that is actually factually incorrect or fabricated✓
• d.It refuses outright to answer any question that happens to involve numbers or arithmetic

A hallucination is generated content that sounds plausible and confident but is factually wrong or invented. It is a core reliability risk of LLMs and a reason to ground outputs in verified sources and to review them.

A company wants a foundation model to answer questions using its own internal documents without retraining the model. Which technique fits best?

• a.Removing the guiding system prompt so the model answers freely
• b.Retrieval Augmented Generation, which fetches relevant documents and adds them to the prompt✓
• c.Training an entirely brand-new foundation model completely from scratch on those same documents
• d.Increasing the model's temperature setting all the way up to a very high randomness value

RAG retrieves relevant passages from a knowledge source at query time and injects them into the prompt, grounding answers in current, proprietary data without the cost of retraining. Higher temperature only adds randomness.

What is the most likely effect of setting a very high temperature value when generating text from an LLM?

• a.The output turns strictly deterministic and comes out identical on every single run without variation
• b.The model begins to automatically fact-check every sentence it writes
• c.The model quietly ignores the entire prompt and returns an empty response
• d.The generated output becomes more random, more varied, and generally more creative✓

Temperature controls randomness in token sampling. Higher temperature yields more diverse, creative, less predictable text; lower temperature makes output more focused and repeatable. It does not verify facts.

Which of the following is a realistic and appropriate use case for generative AI?

• a.Serving as the single, completely unreviewed final authority for patient medical diagnoses
• b.Drafting a first version of marketing copy that a human then reviews and edits✓
• c.Fully replacing every relational and object data storage system across the whole organization
• d.Guaranteeing perfectly accurate binding legal rulings with absolutely no human oversight at all

Generative AI is well suited to drafting content that humans review, such as marketing copy. In high-stakes domains like medicine or law it should assist experts, not replace their judgment, because of hallucination and accountability risks.

What is an embedding in the context of generative AI and semantic search?

• a.A numeric vector representation of data that captures its underlying semantic meaning✓
• b.The specific physical AWS Region and data-center server location where a model is hosted
• c.A hard-coded list of banned words the model must avoid
• d.A heavily compressed image file used to reduce storage cost

Embeddings map text (or images, audio) into numeric vectors so that items with similar meaning sit close together in vector space. They power semantic search, clustering, and RAG retrieval.

How does a general-purpose foundation model differ from a narrow, task-specific traditional ML model?

• a.It can handle only one single narrow task, unlike the more flexible traditional model
• b.It requires no training data , while a traditional model needs data
• c.It is genuinely identical in scope and differs only in name
• d.It is broadly capable and adaptable to many tasks, not built for a single one✓

Foundation models are trained broadly and adapt to many downstream tasks with prompting or light tuning. Traditional ML models are typically trained for one specific task and do not generalize beyond it.

A developer wants a coding assistant integrated with AWS that can answer questions and help write and debug code. Which service is designed for this?

• a.Amazon Q Developer, the generative AI assistant built for coding and AWS tasks✓
• b.Amazon Forecast, the fully managed service that predicts future numeric values from historical time-series data
• c.Amazon Kendra, the enterprise search service
• d.Amazon Transcribe, the service that converts speech into text

Amazon Q Developer is the generative AI assistant for software development and AWS tasks, helping write, explain, and debug code. Kendra is enterprise search, Transcribe is speech-to-text, and Forecast handles time-series prediction.

Which factor should most influence the choice of one foundation model over another for a given application?

• a.Only the alphabetical order of the model provider's name
• b.Whichever particular model happened to be released most recently, ignoring every other factor
• c.The trade-offs among capability, latency, cost, and fit for the specific task✓
• d.Always picking the model with the very largest parameter count regardless of need

Model selection balances task fit, output quality, latency, and cost. The largest or newest model is not automatically best; a smaller, cheaper model may meet requirements with lower cost and faster responses.

What is a prompt in the context of a large language model?

• a.The input text or instructions given to the model to guide the response it generates✓
• b.The lossless compression algorithm that is used to store the model's weights on disk
• c.The graphics processing unit hardware that actually runs the model's heavy computations
• d.The billing invoice document that is automatically generated after every single API call

A prompt is the input, the instructions, context, and questions, that you provide to steer the model's output. Crafting effective prompts (prompt engineering) strongly shapes the quality and relevance of responses.

Which description best defines prompt engineering?

• a.Physically upgrading the servers on which the model runs
• b.Manually relabeling the model's entire training dataset
• c.Designing and refining the input instructions to a model to get better outputs✓
• d.Rewriting the model's internal source code in order to change its neural architecture

Prompt engineering is the practice of crafting and refining prompts, wording, context, examples, and structure, to guide a model toward higher-quality outputs, without altering the model's weights or code.

A prompt includes two or three worked examples before asking the model to handle a new case. What prompting technique is this?

• a.Fine-tuning, which permanently updates the model's internal weights using a labeled dataset
• b.Zero-shot prompting, which supplies the model with no examples
• c.Data augmentation, which synthetically expands the training set by transforming existing samples
• d.Few-shot prompting, which supplies a handful of example input-output pairs as guidance✓

Few-shot prompting includes a small number of demonstrations in the prompt so the model can infer the desired pattern. Zero-shot gives no examples. Neither changes the model's weights, unlike fine-tuning.

When is fine-tuning a foundation model more appropriate than prompt engineering alone?

• a.When you need results instantly, with zero setup time and no additional training cost whatsoever
• b.When you specifically want to avoid using any of your own organization's data at all
• c.When you only need to slightly change the wording of one single question you are asking
• d.When the model must reliably adopt specialized knowledge or a style prompting cannot achieve✓

Fine-tuning adjusts model weights on your curated data, useful for deeply embedding domain knowledge, tone, or formats that prompting cannot reliably produce. It costs more and takes longer than simply editing a prompt.

In Amazon Bedrock, what does a Knowledge Base primarily enable?

• a.Automatically writing unit tests for your application code
• b.Encrypting all of the network traffic that flows between two separate AWS Regions in transit
• c.Provisioning virtual servers, load balancers, and networking for a scalable web application tier
• d.Connecting foundation models to your own data sources to support retrieval-augmented generation✓

Bedrock Knowledge Bases let you ground foundation models in your proprietary data by managing ingestion, embedding, and retrieval for RAG, so responses can cite current, organization-specific information.

What is an agent in the context of Amazon Bedrock?

• a.A component that lets a model plan and take multi-step actions by calling APIs and tools✓
• b.A physical networking appliance that is installed inside an on-premises corporate data center
• c.A human customer-support representative who manually answers and resolves customer help tickets
• d.A billing alarm that sends you a notification whenever monthly account spending exceeds a limit

Bedrock Agents orchestrate multi-step tasks: the model can break a request into steps, call APIs or tools, and use knowledge bases to complete an objective, going beyond a single text response.

Which AWS AI service extracts printed and handwritten text, forms, and tables from scanned documents?

• a.Amazon Textract, which reads text, form fields, and tables out of documents✓
• b.Amazon Polly, which converts written input text into natural-sounding spoken audio
• c.Amazon Translate, which converts written text from one human language into another
• d.Amazon Lex, which builds conversational chatbots from speech and text

Amazon Textract performs OCR-plus: it extracts text, key-value form fields, and tables from documents and images. Polly is text-to-speech, Lex builds chatbots, and Translate handles language translation.

A company wants to convert written text into natural-sounding speech for an accessibility feature. Which service fits?

• a.Amazon Comprehend, which analyzes the meaning and sentiment of text
• b.Amazon Transcribe, which converts recorded or streaming speech into written text
• c.Amazon Rekognition, which analyzes objects and scenes in images and video
• d.Amazon Polly, which synthesizes lifelike spoken audio from written text✓

Amazon Polly is text-to-speech, turning written text into lifelike audio. Transcribe does the reverse (speech-to-text), Comprehend analyzes text meaning, and Rekognition analyzes images and video.

Which service automatically detects the dominant language, key phrases, entities, and sentiment in a body of text?

• a.Amazon Textract, a service that extracts written content out of scanned documents
• b.Amazon Polly, a service that turns written text into natural-sounding spoken audio
• c.Amazon Kendra, a service that provides intelligent search across content
• d.Amazon Comprehend, a natural language processing service for analyzing text✓

Amazon Comprehend is a natural language processing service that identifies language, entities, key phrases, sentiment, and more within text. Textract extracts document content, Polly is TTS, and Kendra is enterprise search.

A business wants to build a conversational chatbot with automatic speech recognition and natural language understanding. Which service is purpose-built for this?

• a.Amazon Polly, which converts written input text into natural-sounding spoken audio
• b.Amazon Textract, which pulls text, forms, and tables from scanned documents
• c.Amazon Forecast, which predicts future values from historical time-series input data
• d.Amazon Lex, which supplies speech recognition and language understanding for bots✓

Amazon Lex provides the automatic speech recognition and natural language understanding needed to build voice and text chatbots. Forecast handles time-series, Textract reads documents, and Polly synthesizes speech.

Which service provides intelligent, natural-language enterprise search across an organization's internal content repositories?

• a.Amazon Translate, which converts written text from one human language into another one
• b.Amazon Rekognition, which detects objects, scenes, and faces within images and video
• c.Amazon Kendra, which understands natural-language queries across internal content✓
• d.Amazon Transcribe, which converts recorded or streaming speech audio into written text

Amazon Kendra is an intelligent enterprise search service that understands natural-language queries across documents and data sources. The other services handle images, speech-to-text, and translation respectively.

Which AWS service analyzes images and video to detect objects, scenes, faces, and unsafe content?

• a.Amazon Polly, the service that converts written text into natural-sounding speech
• b.Amazon Comprehend, the natural language service for analyzing text
• c.Amazon Rekognition, the computer vision service for analyzing images and video✓
• d.Amazon Lex, the service for building chatbots

Amazon Rekognition is the computer vision service for images and video: object and scene detection, facial analysis, text-in-image, and content moderation. Comprehend works on text, not visuals.

What is the main advantage of using a managed AWS AI service like Amazon Rekognition instead of building a custom model?

• a.You are fully guaranteed that the model will never make a mistake
• b.You get production-ready capabilities through an API with no data, training, or infrastructure work✓
• c.You are given complete low-level control over the model's internals
• d.You must still collect and label millions of your own images first

Managed AI services expose ready-to-use capabilities through simple APIs, removing the burden of data collection, training, tuning, and infrastructure. The trade-off is less control over the underlying model internals.

In a RAG-based application, what is the typical role of a vector database?

• a.Scheduling and running the nightly backups of the database
• b.Compiling and bundling all of the application's front-end JavaScript and CSS source code
• c.Storing embeddings so relevant content can be retrieved quickly to augment the prompt✓
• d.Rotating the cryptographic encryption keys that are used to protect stored data at rest

A vector database stores embeddings and supports fast similarity search, so the most semantically relevant chunks can be retrieved and added to the model's prompt. This retrieval step is the 'R' in RAG.

Why might adding clear context and explicit constraints to a prompt improve a model's output?

• a.It fully disables the model's ability to make a factual mistake
• b.It reduces ambiguity, helping the response better match the user's intent and format✓
• c.It reliably lowers the price that is charged per token to zero
• d.It permanently retrains the underlying model on the instructions

Specific context, role, and constraints narrow the space of acceptable answers, guiding the model toward the intended content, tone, and format. It shapes a single response and does not retrain the model or change pricing rules.

A team wants to customize a foundation model's behavior using labeled examples, accepting higher cost than prompting. Which approach is this?

• a.Reducing the maximum output token limit for each response
• b.Simply lowering the temperature sampling parameter at inference time
• c.Fine-tuning the model on their own curated dataset of labeled example pairs✓
• d.Switching the language of the final output by routing it through a translation service

Fine-tuning updates model weights using labeled examples to adapt behavior, at greater cost and effort than prompt engineering. Adjusting temperature or token limits only tweaks inference settings, not learned behavior.

What does bias in a machine learning model most commonly refer to?

• a.The total amount of electrical power that is consumed by the hardware during model training
• b.The alphabetical ordering of the feature columns as they appear within the source dataset
• c.Systematic errors causing unfair, skewed outcomes, often from imbalanced training data✓
• d.The random noise in predictions that reliably averages out to zero over many repeated trials

Bias is systematic, unfair skew in a model's outcomes, frequently traceable to unrepresentative or imbalanced training data. It can disadvantage certain groups and is a central concern of responsible AI.