Iterative Refinement Techniques

In Module 3, we learned the loop. Now, we learn the Advanced Moves within that loop.

1. The "Critique-and-Revise" Pattern

Ask the AI to perform a "post-mortem" on its own work.

Example Prompt:

1. "Write a blog post about [Topic]."
2. (AI responds)
3. "Critique the post you just wrote. Is it too wordy? Is the hook strong enough? List 3 specific improvements."
4. "Now, rewrite the post incorporating those 3 improvements."

2. Layering Constraints

Adding one constraint at a time is often better than all at once.

• Step 1: Write the logic.
• Step 2: "Now, adapt the tone to be professional."
• Step 3: "Finally, format it for a mobile screen with short paragraphs."

graph TD
    Base[Base Output] --> Critique[Self-Critique]
    Critique --> Polish[Apply Polish]
    Polish --> Format[Final Formatting]
    Format --> Result[Exceptional Result]

3. The "Variational" Prompt

Ask for 3 different versions and pick the best elements from each.

• "Write 3 different opening hooks for this article. One should be a question, one should be a shocking fact, and one should be a personal story."

4. The "Tone Mirror" Technique

• "Here is a paragraph I wrote. Identify the tone and vocabulary level. Now, rewrite the following [Other Text] to match that exact style."

---

Hands-on: The 3-Step Polish

1. Prompt: "Explain how a plane flies."
2. Refinement 1: "Make it sound like a 1950s detective noir novel."
3. Refinement 2: "Now, make it under 100 words and add 2 emojis."

Key Takeaways

• Don't settle for the first response.
• Use Self-Critique to find flaws you might miss.
• Layering prevents the model from being overwhelmed.