Skip to main content

Challenge 19: Generative AI Fundamentals

Estimated Time

20-30 min | Cost: Free | Domain: Generative AI (15-20%)

Exam skills covered

  • Identify features of generative AI models
  • Identify common scenarios for generative AI
  • Identify capabilities of Azure OpenAI Service

Overview

Generative AI refers to artificial intelligence models that can create new content — text, images, code, audio, and video — based on patterns learned from vast amounts of training data. Unlike traditional AI that classifies or predicts, generative AI produces original outputs that didn't exist before. When you ask a chatbot to write an email, generate an image from a description, or complete your code, you're using generative AI.

The backbone of modern text-based generative AI is the Large Language Model (LLM). LLMs like GPT-4 are trained on enormous datasets of text from the internet, books, and other sources. They learn patterns, grammar, facts, and reasoning abilities — then generate new text by predicting what comes next, token by token. They don't "understand" in the human sense, but they're remarkably capable at producing coherent, contextual responses.

Foundation models are large, pre-trained models designed to be adapted for many tasks. Rather than training a model from scratch for each use case, organizations can take a foundation model (like GPT-4) and fine-tune it or prompt-engineer it for specific needs. This "train once, use everywhere" approach has revolutionized AI development — making powerful AI accessible without requiring massive compute budgets or datasets.

Explore

Task 1: Understand what makes AI "generative"

Compare traditional AI vs. generative AI:

AspectTraditional AIGenerative AI
What it doesClassifies, predicts, detectsCreates new content
OutputLabel, number, or categoryText, images, code, audio
Example"This email is spam""Write a professional reply to this email"
Training approachTask-specific datasetsMassive diverse datasets
AdaptabilityOne task per modelMany tasks per model

Types of generative AI models:

Model TypeWhat it generatesExamples
Large Language Models (LLMs)Text, code, reasoningGPT-4, GPT-4o, GPT-3.5-Turbo
Diffusion modelsImages from text descriptionsDALL-E, Stable Diffusion
Audio modelsSpeech, music, transcriptionWhisper (transcription)
Multimodal modelsText + images combinedGPT-4o (vision + text)

Task 2: Explore common generative AI scenarios

Generative AI enables many practical business scenarios:

ScenarioDescriptionExample
Content creationGenerate text, summaries, reportsMarketing copy, blog posts
Code generationWrite, explain, or debug codeGitHub Copilot, code completion
Conversational AINatural dialogue with usersCustomer service chatbots
Image generationCreate images from text descriptionsProduct mockups, design concepts
Document summarizationCondense long documentsMeeting notes, research papers
Data analysisInterpret data and generate insightsNatural language queries on databases
Translation & localizationContextual translation beyond word-by-wordMarketing content adaptation

Task 3: Try generative AI in action

Try Microsoft Copilot (free): copilot.microsoft.com

  1. Text generation: Ask "Write a short professional email declining a meeting invitation politely"
  2. Summarization: Paste a long paragraph and ask "Summarize this in 2 sentences"
  3. Code generation: Ask "Write a Python function that calculates the factorial of a number"
  4. Creative content: Ask "Write a haiku about cloud computing"
  5. Reasoning: Ask "If a train travels at 60 mph for 2.5 hours, how far does it go? Show your work."

Observe how the model:

  • Generates coherent, contextual responses
  • Follows instructions and format requests
  • Adapts tone based on your prompt
  • Can handle diverse tasks with a single model

Task 4: Understand key generative AI concepts

Review these fundamental concepts that appear on the exam:

ConceptDefinition
TokenThe basic unit of text processing — roughly ¾ of a word. "Hello world" ≈ 2 tokens
Context windowMaximum tokens a model can process (input + output combined)
TemperatureControls randomness: 0 = deterministic/focused, 1 = creative/varied
Top-pAlternative randomness control: limits token selection to most probable options
PromptThe input/instruction you give the model
CompletionThe output/response the model generates
GroundingConnecting model output to real, verifiable data sources
HallucinationWhen the model generates plausible but factually incorrect information

Temperature example:

  • Temperature 0: "The capital of France is Paris." (always the same)
  • Temperature 1: "The capital of France is Paris, a city known for..." (may vary each time, more creative)
tip

The exam expects you to know that generative AI can produce incorrect information (hallucinations) and that techniques like grounding help mitigate this. This connects to responsible AI principles.

Key Concepts

ConceptDefinition
Generative AIAI models that create new content (text, images, code) rather than just classifying or predicting
Large Language Model (LLM)AI model trained on massive text data that generates text by predicting the next token
Foundation modelA large pre-trained model designed to be adapted for many downstream tasks
TokenThe basic unit of text for LLMs — approximately ¾ of a word
HallucinationWhen an AI model generates plausible-sounding but factually incorrect content
GroundingConnecting AI outputs to real data sources to improve accuracy and reduce hallucinations

Common Misconceptions

MisconceptionReality
Generative AI "understands" like humans doLLMs predict probable next tokens based on patterns — they don't have true comprehension or consciousness
Generative AI always produces correct informationModels can hallucinate — generating confident but incorrect statements. Always verify critical information
You need to train your own model to use generative AIFoundation models are pre-trained and ready to use; you can prompt-engineer or fine-tune them
Generative AI replaces all other AITraditional AI (classification, detection, prediction) remains better for many structured tasks
Higher temperature always means better outputHigher temperature increases creativity but also increases randomness and potential for errors

Knowledge Check

1. What distinguishes generative AI from traditional AI approaches?

2. What is a "hallucination" in the context of generative AI?

3. What does the "temperature" parameter control in a generative AI model?

4. What is a foundation model?

5. Which of the following is a common scenario for generative AI?

Learn More