Cognitive Robotics

Explain Like I’m 12 — How to Talk to AI and Actually Get What You Want

A friendly, professional guide to prompting AIs: why it felt sudden, common prompt mistakes, a plain-language peek under the hood, and practical prompt recipes (useful, not magical).

ELI12 style

Prompt recipes

Plain-English

TLTL;DR

AI is best treated like a conversational assistant: be clear, set a role, pick a format, and give examples. Use “Explain like I’m 12” to force simplicity. If the first answer isn't great, iterate — that's how you get reliably useful output.

1From switches to conversations: the human ↔ machine shift

We used to instruct machines with buttons, scripts, and rigid interfaces. Today we often talk to them. That change — from commands to conversation — is more than cosmetics: it changes expectations. A map app returns routes; an AI assistant can weigh tradeoffs, ask follow-ups, and propose alternatives based on your preferences. Conversational AI feels like a helper because it keeps context and responds in natural language. The better you explain what you want, the more human-like and useful the result becomes.

2Why the AI leap felt sudden

That “sudden” feeling came from many small improvements compounding: larger models, more data, faster chips, and better training tricks. When models cross certain size or data thresholds they sometimes unlock new capabilities quickly — so the progress looks abrupt from the outside. Think of it like turning up the resolution: a little higher might not matter; a lot higher suddenly reveals dramatic detail.

3Common prompts people use — and why ELI12 helps

Most prompts fall into a few categories: quick Q&A, rewrite/summarize, role-play, step-by-step plans, or creative generation. The common failure modes are vagueness, missing context, hallucinated facts, wrong audience level, and unclear format.

Quick rule: If your prompt is vague, expect a vague answer. Tell the AI who it should be and how to respond.

Using “Explain like I’m 12” (ELI12) is an easy way to ask the AI to simplify. It signals: no jargon, short explanations, and friendly analogies. Example:

Before

What is AI?

The model might answer with a long, technical definition full of jargon.

After (ELI12)

Explain AI like I’m 12.

“Think of AI as a very smart helper that learned from reading lots of books. When you ask it a question, it uses what it learned to give a friendly answer.”

4How AI “thinks” — plain language mechanics

AI models don’t “think” like humans — they do fast statistical pattern matching. Here are the key ideas in simple terms.

Tokens — the Lego bricks of language

Text is broken into tokens: pieces of words or punctuation. The model predicts one token at a time to build an answer.

Embeddings — coordinates on a map

Each token becomes a list of numbers (an embedding). Similar meanings live near each other on that numeric map — like clustering friends together on a playground map.

Attention — a flashlight for context

Transformers use attention to decide which words in the input matter most for predicting the next token. Imagine each word asking the others, “Are you important for my sentence?” The model uses those signals to focus its prediction.

Next-token prediction & decoding

The trained model scores possible next tokens and picks one based on probabilities. How it chooses (greedy vs. sampling) affects whether the output is conservative or creative.

Analogy: The whole process is like a supercharged autocomplete that uses a big map of language and a spotlight to pick the most fitting next word.

5How to improve your prompts — recipes that work

Want reliable, useful answers? Use these prompt-building blocks.

Role / context

Tell the AI who it should be. Examples: “You are a friendly science tutor,” or “You are a senior Python developer.” This shapes tone and depth.

Audience & level

Say who the explanation is for: “Explain to a 12-year-old,” “Explain to a product manager,” or “Explain to a beginner.”

Format & constraints

Ask for structure: “Give 3 bullet points,” “Keep it under 120 words,” or “Provide a one-line summary and a code snippet.” Specify length and style to control output shape.

Examples (few-shot)

Show one or two ideal outputs and ask the model to follow that pattern. “Here’s how I want the answer formatted — please match this.”

Ask for assumptions & sources

Request the model list its assumptions or cite where applicable. Ask it to say “I don’t know” when unsure.

Iterate

Use follow-up prompts to refine. Start coarse, then ask for simplification, examples, or constraints. Treat the conversation as an iterative workshop.

Prompt template — Explain (ELI12)

You are a friendly teacher. Explain <TOPIC> to a 12-year-old in three bullets and give one everyday analogy.

Prompt template — Debug code

You are an expert Python developer. Here is my code: <PASTE>. 1) One-sentence diagnosis. 2) Show corrected code (only changed lines). 3) List 2 tests to verify the fix.

Before → After example

Before: Tell me about climate change.
After: You are an environmental scientist. Explain climate change in three simple bullets to a 12-year-old and give one real-world example.

6Hands-on lab — three tiny exercises to try

1. Explain — Prompt: Explain blockchain like I’m 12, in three bullets and one analogy. Check: are bullets short and the analogy clear?
2. Debug — Provide a short buggy Python snippet and use the Debug template above. Check: did the model point out the real bug and give a minimal fix?
3. Plan — Prompt: You are a pragmatic project manager. Create a 3-step plan to build a simple Raspberry Pi camera project and list one dependency per step. Check: are steps actionable and dependencies realistic?

These small exercises build intuition quickly: compare the first answer to improved answers after adding role, audience, and format constraints.

FFAQ & Glossary

FAQ — quick answers

• Why does the model make stuff up? It predicts plausible continuations; sometimes plausible is wrong — that’s a hallucination.
• How do I stop hallucinations? Ask for sources, ask the model to say “I don’t know,” and verify facts independently.
• What is a token? A small piece of text the model processes (a word or part of a word).
• Should I trust AI for facts? Not without verification — use it for drafts, ideas, and explanations, but check critical facts.

Glossary (short)

• Token: a text fragment the model handles.
• Embedding: numeric representation of meaning.
• Attention: how the model decides what parts of text to focus on.
• Hallucination: produced but incorrect or fabricated information.

✓Conclusion & Practical Takeaways

Talking to AI is a skill, not sorcery. Use role + audience + format, give an example, and iterate. When in doubt, ask the model to simplify — “Explain like I’m 12” reliably forces clarity. Use AI for brainstorming, drafting, and explanations, but always verify anything that matters.

Set a role and audience

Specify format and length

Show one example output

Ask for assumptions or sources

Try this starter prompt: You are a friendly teacher. Explain <TOPIC> to a 12-year-old in three bullets and give one everyday analogy.

Liked this guide? Try the exercises above and paste your best improved prompt in the comments. Want a printable cheat-sheet or prompt templates? Add a quick note and I’ll include them in the next post.