Microsoft · AI-500
Advanced prompt engineering and fine-tuning
AI-500 · Lesson 5 of 16 · 40:19
YouTube loads only after you press play. This chapter opens at its timestamp in the full course. Watch on YouTube ↗
In this lesson
Prepare for the AI-500 exam (Microsoft Certified: Multi-Agent AI Solutions Expert). In this lesson: few-shot examples, clear syntax and structured outputs, dynamic context injection and context-aware behaviors, defensive guidelines against direct and indirect prompt injection and prompt leakage (Prompt Shields, output validation, persona stability), prompt lifecycle management with versions, A/B tests, and agentic evaluators, and a fine-tuning strategy: when to fine-tune, SFT, DPO, or RFT, training data, and retraining frequency. Includes three original practice questions.
Preview the topics
- Contoso’s visa agent reads PDFs that travelers upload, and one of them hid white text telling the agent to approve the visa. Which defense targets this attack first?
- A Contoso developer edits the refund agent’s prompt directly in production, and nobody can tell which change caused the new errors. What should the team adopt?
- Contoso’s labeling agent needs a strict JSON format and specialized visa vocabulary, the team has 2,000 reviewed examples, and its instructions keep growing. What should the team consider?