Microsoft Azure AI Fundamentals Practice Test

Explanation

• Azure OpenAI Service provides access to different AI models, each specialized for certain tasks:
  • GPT-3.5 / GPT-4 ? Text generation, chat, reasoning, code completion.
  • DALL-E ? Image generation from text prompts.
• Since your requirement is to create an original image (not text), you need an image generation model. That’s exactly what DALL-E is designed for.
• Option A (GPT-3.5 Turbo) is incorrect because GPT models generate text, not images.
• Option B (Bing search) is incorrect because searching and downloading images does not create an original image; it only retrieves existing ones, which may have copyright issues.
• Option C (GitHub Copilot) is incorrect because Copilot is a code assistant, not an image generation tool.
• Option D (DALL-E) is correct because it allows you to generate original, AI-created images from prompts, which fits your scenario.

Explanation

• GPT models (like GPT?3.5 or GPT?4 in Azure OpenAI) are designed for natural language processing tasks, including:
  • Summarization of text
  • Translation
  • Question answering
  • Content generation
• Whisper ? Used for speech-to-text transcription (audio ? text). Not suitable for summarizing documents.
• DALL?E ? Used for image generation from text prompts. Not suitable for text summarization.
• Codex ? Specialized for code generation and programming tasks. Not intended for summarizing general text.
• GPT ? The correct choice because it can read and understand text, then produce a concise summary.

Explanation:

• Content filtering is the mechanism used in Azure OpenAI to detect and block harmful, hateful, or unsafe outputs before they are returned to the user.
• It works by scanning the generated response against safety categories (e.g., hate speech, violence, sexual content) and preventing unsafe content from being delivered.

? Why not the others?

• A. Abuse monitoring ? Tracks usage patterns and detects misuse, but it doesn’t directly stop hateful responses from being generated.
• C. Fine-tuning ? Customizes a model for specific tasks or domains, but it doesn’t guarantee prevention of harmful outputs.
• D. Prompt engineering ? Helps guide the model toward desired behavior, but it’s not a reliable safeguard against hateful responses

Explanation

• A. Extract the invoice number from an invoice ? ? Correct. Document Intelligence can parse invoices and pull out key fields like invoice number, date, total, etc.
• B. Identify the retailer from a receipt ? ? Correct. It can analyze receipts and extract merchant information, transaction details, and amounts.
• C. Find images of products in a catalog ? ? Incorrect. Document Intelligence focuses on text and structured data extraction, not image recognition.
• D. Translate a form from French to English ? ? Incorrect. Translation is handled by Azure Translator, not Document Intelligence.

Explanation:

• Azure AI Vision provides face detection and recognition capabilities, including:
  • Detecting faces in images.
  • Identifying and verifying whether two faces belong to the same person.
  • Finding similar faces across a set of images by comparing facial features.
• B. Azure AI Custom Vision ? Used for training custom image classification or object detection models. It’s great for identifying custom objects (e.g., logos, products), but not for face similarity.
• C. Azure AI Language ? Focused on text analysis (sentiment, translation, summarization, etc.), not images.
• D. Azure OpenAI Service ? Provides GPT, DALL?E, Whisper, and Codex models for text, image generation, and speech-to-text, but not face recognition.