6.3. Reflection Checkpoint: Generative AI Mastery

What breaks without mastering this content? Generative AI makes up 20-25% of questions—the largest domain. Confusion between model capabilities leads to wrong answers. Imagine seeing "create marketing images from descriptions" and not instantly recognizing that's DALL-E—not GPT, not Embeddings.

Consider these questions like capability boundary tests. For instance, the DALL-E question trips up many people—DALL-E GENERATES images but CANNOT describe or analyze them. Image analysis requires a different model. If you hesitate on content filter layers versus metaprompt layers, review Section 6.1.3.

1. Which generative AI model creates images from natural language prompts?

   • DALL-E. GPT models generate text, Embeddings create numerical vectors, Whisper transcribes speech.

2. At which layer are content filters applied in Microsoft's responsible AI model?

   • Safety System layer. This layer provides platform-level content filtering.

3. What can system messages be used to identify?

   • Constraints and styles for generative AI model responses. They set behavioral expectations.

4. Which capability is NOT supported by DALL-E?

   • Image description/analysis. DALL-E generates images but cannot analyze them.

5. What can Embeddings be used for?

   • Searching, classifying, and comparing sources of text for similarity.

6. What uses plugins to provide end users with the ability to get help from a generative AI model?

   • Copilots. They integrate AI assistance into applications.