Show, don’t tell | Master AI Automation in 4 hours Master AI Automation in 4 hours Course About Ayush Modules Sample chapter Toolbox The Microcap Minute Classroom / Module 02: Talking to Machines: Prompting / Chapter 3 Show, don’t tell Watch first, then read. Same lesson, your pace. What you will learn – Few-shot prompting: teaching by example – Constraints that actually constrain – The reverse prompt: making AI ask you questions Examples beat adjectives “Write it in a fun, casual tone”, what does that even mean? Fun like a birthday card, or fun like a stand-up set? Instead of describing the style, show a sample : Write my product descriptions in this exact style: Example: “This bottle keeps chai hot for 8 hours. Tested on a Delhi winter. You’re welcome.” One good example communicates tone, rhythm, attitude and length faster than ten adjectives. Two or three examples teach pattern better still. This technique is called few-shot prompting : zero examples = zero-shot (pure description), one example = one-shot, a few = few-shot. The machine is a pattern-completion engine (Module 1), so feed it the pattern. Where this shines: – Consistent voice across many pieces of writing – Data formats : show one filled row, get twenty matching ones – Difficulty levels : show a solved problem, ask for similar ones Constraints that actually constrain Vague constraint: “keep it short”. Real constraints have numbers and teeth: Weak Strong Keep it short Under 80 words Make it simple Explain to a 12-year-old; no word longer than “electricity” Don’t add fluff No sentence without a fact in it; no opening pleasantries Use my data only Use ONLY the text below. If something isn’t there, write “not stated” That last one deserves attention. “If X is not present, say ‘not stated’” is the single best anti-hallucination line in prompting (Module 1, Chapter 3). It gives the model permission to admit ignorance, most never think to grant it. The reverse prompt: let it interview you Here’s a trick that feels like magic and is just logic. Instead of you guessing what details matter, flip the table: Before answering, ask me up to five questions whose answers would most improve your output. Wait for my answers. Now the model interrogates you , audience? length? deadline? tone?, and you answer casually. Your half-formed request becomes a full specification in two messages. Use it whenever you feel that “I don’t know exactly what I want” fog. Try it yourself Task: get an AI to write your next three social-media captions (or diary headers, or to-do list intros, any repeated format). First ask with pure adjectives. Then give two real examples of lines you love + the constraint “max 15 words, no hashtags unless asked”. Compare. Finally run the reverse prompt on a task you’ve been avoiding because you couldn’t specify it. Save the winner to prompts/ . Key takeaways – One real example teaches more than ten adjectives, use few-shot for voice and formats. – Strong constraints are numeric and specific; always allow “not stated”. – The reverse prompt turns vague requests into complete specs via AI’s own questions. Download the exercise sheet (PDF) Module workbook (PDF) ← Prev: The R-C-T-F recipe Next: The conversation is the product → Classroom / Module 02: Talking to Machines: Prompting / Chapter 3 Show, don’t tell What you will learn – Few-shot prompting: teaching by example – Constraints that actually constrain – The reverse prompt: making AI ask you questions Examples beat adjectives “Write it in a fun, casual tone”, what does that even mean? Fun like a birthday card, or fun like a stand-up set? Instead of describing the style, show a sample : Write my product descriptions in this exact style: Example: “This bottle keeps chai hot for 8 hours. Tested on a Delhi winter. You’re welcome.” One good example communicates tone, rhythm, attitude and length faster than ten adjectives. Two or three examples teach pattern better still. This technique is called few-shot prompting : zer
Show, don’t tell
Written by
in