AWS Certified AI Practitioner (AIF-C01) Practice Exam

Correct: A — supervised learning. Supervised learning trains on labeled examples (input paired with the correct output), so the model learns to predict the label for new inputs — the setup described.

Why not the others: unsupervised learning (B) finds structure in unlabeled data; reinforcement learning (C) learns from rewards through trial and error; a rules engine (D) is not machine learning at all.

Correct: B — a large model pre-trained on broad data, adaptable to many tasks. Foundation models are trained on vast, general datasets and can then be adapted (through prompting, RAG, or fine-tuning) to a wide range of downstream tasks — the basis of modern generative AI.

Why not the others: a single-task model (A) is the opposite of a foundation model; a foundation model is software, not a physical server (C); and it is a model type, not a billing tier (D).

Correct: C — Amazon Bedrock. Bedrock is AWS’s fully managed service for building generative-AI applications with a choice of foundation models through one API, with no infrastructure to manage — exactly the described need.

Why not the others: EC2 Auto Scaling (A) scales virtual servers; S3 Glacier (B) is cold storage; CloudTrail (D) logs API activity. None is a managed generative-AI model service.

Correct: D — retrieval-augmented generation (RAG). RAG retrieves relevant passages from an external knowledge source (such as the company’s documents) and supplies them to the model as context at query time, grounding answers in current, private data without retraining.

Why not the others: temperature (A) affects randomness, not knowledge; a lower token limit (B) truncates output; a smaller model (C) does not add the missing documents. None supplies external context.

Correct: A — prompt engineering. Prompt engineering shapes a model’s output through the wording, examples, and format of the input prompt, without altering the underlying model — the lowest-effort way to steer a foundation model.

Why not the others: retraining (B) and hyperparameter tuning (D) change the model itself and require far more effort and data; data labeling (C) prepares training data, not prompts.

Correct: B — fine-tuning on domain-specific data. Fine-tuning continues training a foundation model on a labeled, domain-specific dataset, adjusting its weights so it specializes — the described legal-language use case.

Why not the others: prompt engineering (A) does not change weights; a larger context window (C) lets the model read more input but does not specialize it; MFA (D) is a security control, unrelated to model behavior.

Correct: C — Amazon SageMaker. SageMaker is AWS’s fully managed service for building, training, and deploying machine-learning models across the whole lifecycle — the standard choice for custom ML.

Why not the others: Polly (A) turns text into speech; Lex (B) builds conversational chatbots; CloudFront (D) is a content delivery network. They are narrow AI/infrastructure services, not a full ML platform.

Correct: D — bias and fairness. Systematic, unjustified favoring of one group is a bias/fairness problem — a core responsible-AI concern that calls for evaluation, mitigation, and monitoring so the model treats groups equitably.

Why not the others: latency (A), storage cost (B), and right-sizing (C) are performance or cost concerns, not fairness issues.

Correct: B — transparency and explainability. Explainability is the ability to understand and communicate how a model reaches its outputs; transparency covers being open about how the system works — both are responsible-AI properties AIF-C01 emphasizes.

Why not the others: elasticity (A) and horizontal scaling (C) are infrastructure scaling concepts; data compression (D) is a storage technique. None concerns understanding a model’s decisions.

Correct: C — managing access to their data and who can use the service. As with all AWS services, the customer is responsible for security in the cloud — their data, identity and access controls, and configuration — while AWS secures the underlying service and facilities.

Why not the others: physical data-center security (A), patching managed infrastructure (B), and the network backbone (D) are all AWS responsibilities under the shared responsibility model.