Sample questions

Applications of Foundation Models

1. What does Retrieval Augmented Generation (RAG) do?

• AIt retrieves relevant external content and supplies it to the model as context.
• BIt permanently rewrites the model's internal weights on every user request.
• CIt compresses the model so that it consumes less memory during inference.
• DIt encrypts the model's responses before they are returned to the caller.

RAG retrieves relevant information from an external source and includes it in the model's context so responses are grounded in that content. It does not modify model weights, compress the model, or handle encryption.

Applications of Foundation Models

2. A company wants its AI assistant to answer questions using the company's own internal policies, which the foundation model was never trained on. Which approach fits best?

• AIncreasing the temperature setting so the model produces more varied answers.
• BRetrieval Augmented Generation, supplying the policy content as retrieved context.
• CReducing the maximum response length so answers are returned more quickly.
• DMoving the application to a different AWS Region closer to the company office.

RAG lets a model answer from content it was never trained on by retrieving that content and placing it in context. Temperature, response length, and Region placement affect variability, speed, and latency rather than knowledge access.

Applications of Foundation Models

3. Which AWS capability provides a managed way to implement Retrieval Augmented Generation against a company's own documents?

• AAmazon CloudFront, which caches and delivers content from edge locations.
• BAWS Budgets, which alerts when spending approaches a defined threshold.
• CAmazon Bedrock Knowledge Bases, which manages retrieval over a company's content.
• DAmazon Route 53, which registers domains and routes DNS queries to endpoints.

Amazon Bedrock Knowledge Bases provides managed retrieval over a customer's own content for RAG workflows. CloudFront caches web content, Budgets handles cost alerts, and Route 53 provides DNS.

Applications of Foundation Models

4. A team wants their model's answers to be more focused and consistent rather than varied and creative. Which inference parameter should they lower?

• AThe storage class, which determines how retrieved documents are archived.
• BThe instance count, which determines how many servers handle the traffic.
• CThe retention period, which determines how long logs are kept before deletion.
• DThe temperature, which controls how much randomness appears in the output.

Temperature controls output randomness: lower values produce more focused, deterministic responses while higher values increase variety. Storage class, instance count, and retention period are infrastructure settings.

Security, Compliance, and Governance for AI Solutions

5. How does grounding a model's answers in retrieved source content help address hallucination?

• AIt gives the model verified material to base its response upon.
• BIt prevents the model from generating any text in its response.
• CIt automatically corrects factual errors within the source material.
• DIt removes the need for any human review of the generated answers.

Grounding supplies actual source material so responses rest on verified content rather than fabrication. It does not stop generation, fix errors in the sources themselves, or eliminate the value of review.

Applications of Foundation Models

6. Which TWO AWS services can store the embeddings used in a retrieval workflow? (Select TWO.)

• AAmazon OpenSearch Service, which can index and search vector embeddings.
• BAmazon Aurora, which supports storing vector data for similarity search.
• CAmazon Route 53, which registers domain names and routes DNS queries.
• DAWS Artifact, which supplies AWS compliance reports to customers on demand.
• EAmazon CloudWatch, which collects operational metrics and log data.

Amazon OpenSearch Service and Amazon Aurora can both store and search vector embeddings for retrieval workflows, as can Neptune and RDS for PostgreSQL. Route 53, Artifact, and CloudWatch handle DNS, compliance documents, and monitoring.