Curriculum Overview: Generative AI Workloads on Azure (AI-900)
Unit 5: Describe features of generative AI workloads on Azure (20–25%)
Curriculum Overview: Generative AI Workloads on Azure
This curriculum covers Unit 5 of the Microsoft Azure AI Fundamentals (AI-900) certification. This unit represents 20–25% of the total exam weight, making it one of the most significant sections for candidates to master.
Prerequisites
Before diving into Generative AI workloads, learners should ideally possess the following foundational knowledge:
- Basic AI Knowledge: Understanding of common AI workloads like Computer Vision and Natural Language Processing (covered in Units 1–4).
- Cloud Fundamentals: Familiarity with basic Microsoft Azure concepts (Resource Groups, Regions, and Subscription models).
- Machine Learning Basics: A high-level understanding of what "training" a model involves and the concept of a dataset.
- Transformer Architecture: Basic awareness that modern GenAI is built on the Transformer architecture (introduced in Unit 2).
Module Breakdown
| Module | Topic | Complexity | Focus Area |
|---|---|---|---|
| 1 | Foundations of Generative AI | Beginner | LLMs, Tokenization, and Content Generation (Text, Image, Code) |
| 2 | Azure OpenAI Service | Intermediate | GPT-4, DALL-E, and API capabilities |
| 3 | Azure AI Foundry | Intermediate | Model Catalog, Model Benchmarking, and Tooling |
| 4 | Responsible AI for GenAI | Advanced | Mitigation of bias, transparency, and safety filters |
Generative AI Workflow
Learning Objectives per Module
Module 1: Foundations of Generative AI
- Identify features of generative AI models and how they differ from traditional discriminative AI.
- Describe common scenarios for generative AI, including text summarization and creative writing.
- Understand the concept of Tokenization and Embeddings.
Module 2: Azure OpenAI Service
- Describe the features and capabilities of specific models like GPT-4 and DALL-E.
- Identify the unique value proposition of Azure OpenAI compared to public OpenAI (security, privacy, and regional availability).
Module 3: Azure AI Foundry & Model Catalog
- Explain the role of Azure AI Foundry (formerly AI Studio) in unified AI development.
- Explore the Model Catalog to compare and deploy open-source and proprietary models.
Module 4: Responsible AI Considerations
- Identify how the six principles of Responsible AI (Fairness, Reliability, etc.) apply specifically to generated content.
- Describe Azure's built-in safety filters and content moderation tools.
Visualizing Embeddings
Success Metrics
To ensure mastery of this unit, students should aim for the following benchmarks:
- Scenario Recognition: Given a business problem (e.g., "We need to generate marketing emails"), correctly identify the appropriate Azure GenAI service (Azure OpenAI).
- Model Selection: Correcty distinguish when to use GPT (for text/logic) versus DALL-E (for images).
- Responsible AI Integration: Identify which Microsoft tool to use for filtering harmful content in a real-time chat application.
- Practice Exam Performance: Scoring at least 85% on practice questions specifically related to Unit 5.
Real-World Application
Generative AI on Azure is not just a theoretical concept; it is currently transforming industries in the following ways:
- Customer Support: Using GPT models to create intelligent chatbots that provide nuanced, human-like responses to customer queries.
- Software Development: Utilizing GitHub Copilot (powered by Azure OpenAI) to suggest code blocks and debug existing scripts.
- Marketing & Creative: Rapidly generating high-fidelity product images using DALL-E based on simple text descriptions.
- Data Analysis: Summarizing massive document sets into executive summaries or key bullet points automatically.
[!IMPORTANT] The AI-900 exam frequently tests the "Responsible AI" aspect of Generative AI. Always remember that Azure AI services include "Content Filters" by default to maintain safety and compliance.