My default when I need to genuinely understand something new — evals, WhatsApp Business APIs, pricing models — not skim it. Asking for a textbook chapter instead of "explain X" forces depth, structure, and progression. Replace [topic] with yours.
"Teach me [topic] from first principles. Structure it as a 30-page PDF: start with fundamentals, build up to advanced concepts, include examples and diagrams where useful, and end with a summary and further reading. Write it like a textbook chapter, not a listicle."
Condensed sample with [topic] = "evals for AI products" — the real output runs chapter-length:
EVALS FOR AI PRODUCTS — A FIELD GUIDE
1 · Fundamentals. An eval is a repeatable test of model output quality. Where unit tests check code paths, evals check judgment: given this input, is this output acceptable? …
2 · Why traditional QA breaks. The same prompt can produce different outputs; pass/fail becomes a distribution, not a boolean. …
3 · Types of evals. Golden sets, LLM-as-judge, human review panels, A/B in production — with the failure modes of each. …
4 · Building your first eval set. Start from real user transcripts, not imagined cases; 50 well-chosen examples beat 5,000 synthetic ones. …
5 · Advanced: regression gates. Wiring evals into release so a prompt change that hurts quality blocks the ship. …
Summary. Evals are the PM's spec, expressed as tests. Further reading: …
SAMPLE OUTPUT · CONDENSED FOR SPACE