Prompt Design

📌 Learning goals

• Break a vague request into four parts: goal, data, rules, and output.
• Tell the difference between zero-shot, one-shot, and few-shot: the difference is how many examples you give first.
• Know that Chain-of-Thought means working through a problem in steps; it does not mean asking the model to reveal all its private thoughts.
• Use the same small test set (eval) to compare before and after.
• Notice when the problem is not the prompt, then change the model, data, or tool.

📚 Required reading

1. 1.Anthropic Prompt Engineering Tutorial⭐⭐⭐⭐⭐Work through the notebook once.
2. 2.OpenAI Prompt Engineering⭐⭐⭐⭐Read message hierarchy, examples, and evals.
3. 3.Google Prompt Design Strategies⭐⭐⭐⭐See clear instructions, fixed structure, and repeated testing.

🎯 Curated resources

Summaries from the upstream curriculum; full code, cost, and latency estimates live upstream.

1. Exercise 1: the four-part prompt — turn "help me sort this out" into a checkable prompt with goal/data/rules/output.
2. Exercise 2: the power of examples — run the same task zero-shot, one-shot, and few-shot, scoring on a fixed six-question set.
3. Exercise 3: a small eval — change one thing at a time, rerun the same questions, and keep the scores before concluding.
4. The motto: goal → data → rules → output; state it clearly first, decide on examples second, and check with fixed questions last.

✅ Self-check

• I can write out goal, data, rules, and output.
• I can compare before and after using the same six questions.
• I change one thing at a time and keep the scores.
• I know that missing data or needed actions cannot be fixed by prompting alone.