Free AIF-C01 sample questions with answers
Try these before reading the answers. The mix reflects the real weighting, so most come from the generative AI domains.
Question 1. A company wants to predict next quarter's sales figures from five years of historical sales data. Which type of machine learning problem is this?
- A. Classification
- B. Regression
- C. Clustering
- D. Reinforcement learning
Show answer
Answer: B
Predicting a continuous numeric value such as a sales figure is regression. Classification predicts a category, clustering groups similar records without labels, and reinforcement learning trains an agent through rewards. The historical data with known outcomes also makes this supervised learning.
Question 2. A support chatbot built on a foundation model must answer from the company's product manuals, which are updated monthly. Retraining the model every month is not feasible. What should the company implement?
- A. Fine-tune the model each month on the updated manuals
- B. Increase the model’s temperature setting
- C. Retrieval-augmented generation over an index of the manuals
- D. Select a model with a larger context window
Show answer
Answer: C
Retrieval-augmented generation retrieves the current manuals at query time and passes them to the model, so answers stay current without retraining. Fine-tuning bakes knowledge into the model and would need repeating after every update, raising temperature increases randomness, and a larger context window alone does not supply the documents.
Question 3. A company wants access to several foundation models through a single managed AWS service, without provisioning infrastructure. Which service fits?
- A. Amazon Bedrock
- B. Amazon EC2
- C. Amazon S3
- D. AWS Lambda
Show answer
Answer: A
Amazon Bedrock provides managed access to foundation models through one interface with no infrastructure to provision. Amazon EC2 would require managing servers yourself, Amazon S3 is object storage, and AWS Lambda runs code without providing models.
Question 4. A generative AI assistant confidently produces a product specification that does not exist in any company document. What is this behaviour called, and what most directly reduces it?
- A. Overfitting; reduce the number of training epochs
- B. Drift; retrain the model on newer data
- C. Bias; rebalance the training dataset
- D. Hallucination; ground responses in retrieved company content
Show answer
Answer: D
Confident output unsupported by any source is hallucination, and grounding responses in retrieved company content is the most direct mitigation. Increasing temperature makes output more varied rather than more accurate, adding users changes nothing, and switching regions affects where the service runs.
Question 5. Under the AWS shared responsibility model, which is the customer's responsibility when using a managed AI service?
- A. Physical security of the data centre
- B. Managing IAM permissions and protecting the data they submit
- C. Patching the hypervisor underlying the service
- D. Maintaining the managed service’s own infrastructure
Show answer
Answer: B
Customers are responsible for their own data and for controlling who can access it, including IAM permissions, while AWS secures the underlying infrastructure. Physical data centre security, hypervisor patching and the managed service’s own infrastructure are AWS responsibilities.
How did you do?
Notice that none of these required code, mathematics or configuration. That is deliberate and it matches AWS’s published scope: the target candidate uses AI/ML on AWS but does not necessarily build it.
If the generative AI questions felt harder than the fundamentals, that is worth acting on — domains 2 and 3 are 52% of the scored content. See weeks 2 and 3 of the study plan.