Build Generative AI Solutions with Microsoft Foundry AI-102 Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A customer support team uses a prompt flow with a prompt template that includes a few-shot examples. They notice that the model sometimes follows the examples too literally and produces responses that are not appropriate for the current context. They want to maintain the benefits of few-shot learning while reducing overfitting to the examples. What should they do?

   • AAdd a dynamic instruction that tells the model to use the examples only as a style guide, not as a template.
   • BRemove the few-shot examples entirely.
   • CIncrease the number of examples to give the model more patterns.
   • DUse a different model that is less prone to overfitting.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A startup is building a real-time translation feature for a mobile app. They need low latency and support for multiple languages. They are deciding between a large general-purpose model and a smaller specialized translation model. What should they consider?

   • AThe smaller specialized model is likely to offer lower latency and sufficient translation quality.
   • BThey should use the model with the highest benchmark score regardless of latency.
   • CThe larger model is always better for translation quality.
   • DThe model choice does not affect latency.
   Reveal Solution Discussion Select an answer first
4. 3application · medium

   A software company has deployed a prompt flow in Microsoft Foundry and wants to integrate it into their existing .NET application. They need to call the flow from the application and receive the response. Which approach should they use?

   • AUse the Azure CLI to call the flow endpoint from the application.
   • BUse the Microsoft Foundry SDK to connect to the deployed flow endpoint and invoke it from the application.
   • CRe-implement the flow logic directly in the application code.
   • DExport the flow as a Docker container and host it separately from Foundry.
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   Which metric is commonly used to evaluate the relevance of a generative AI model's responses?

   • ACPU utilization
   • BThroughput
   • CGroundedness
   • DLatency
   Reveal Solution Discussion Select an answer first
6. 5expert · hard

   A healthcare organization wants to build a generative AI solution that answers questions from patient education materials. They have strict privacy requirements and need to ensure that the solution does not expose patient data. They also want to minimize the cost of the solution. What is the best approach?

   • AStore the patient education materials in a public blob storage to reduce cost.
   • BFine-tune a model on the patient education materials to improve accuracy.
   • CUse a general-purpose model without grounding to save cost.
   • DUse a RAG pattern with the patient education materials as the grounding data, and ensure the hub is deployed in a compliant region.
   Reveal Solution Discussion Select an answer first