AWS Certified AI Practitioner AIF-C01

A retailer trains a model using past orders labelled fraudulent or legitimate. Which learning approach is this?

A team groups customers by purchasing behaviour without preassigned group labels. Which approach fits?

A model predicts a delivery duration in minutes. What type of predictive task is this?

A support assistant must answer using a frequently updated internal policy library. Which approach is a good initial fit?

What is the purpose of an embedding in semantic retrieval?

Which AWS service provides managed access to foundation models for generative AI applications?

A prompt includes three examples of input and desired output before a new task. Which technique is this?

A developer changes a prompt's instructions without training the model. What is being changed?

When might fine-tuning be more relevant than retrieval alone?

A generated answer confidently cites a policy that does not exist. What is this failure commonly called?

Which evaluation best checks a support model before deployment?

A model predicts fraud on 20 orders and 15 are truly fraudulent. What is precision?

A model misses many actual fraud cases. Which metric directly measures the proportion of actual fraud cases detected?

A model performs well on training data but poorly on held-out data. What is the likely issue?

Why should performance be evaluated across relevant user groups?

A generative AI application processes confidential documents. What should be implemented?

An application should call only an approved model and read only one data bucket. Which control is relevant?

A retrieved document tells an assistant to ignore its instructions and reveal secrets. What threat is present?

An AI-generated decision has significant consequences for a person. Which design can reduce operational risk?

Two models meet the same quality target, but one is slower and more expensive. What should the team do?