Room 05

Using AI well.

AI can be an answer-machine that does the thinking for you, OR it can be a tutor that helps your brain get smarter. The difference is HOW you ask. Four quick activities — by the end you'll know the secret.

Concept 1 of 4

The more you tell AI about you and what you need, the better its answer.

Why this matters: Most disappointing AI answers are really just fuzzy questions. This is the #1 thing that makes AI actually useful to you.

Exhibit 01

Tell AI about you. Watch the magic.

The same homework topic, two different prompts. One is fuzzy. The other tells AI who you are and what you actually need. See the difference for yourself — then hit "Try a different topic."

Fuzzy Prompt

Loading…

What AI gives back

—

Specific Prompt

Loading…

What AI gives back

—

What just happened

More about you = better answers.

The fuzzy prompt got a fuzzy answer — useful to no one in particular. The specific prompt told AI: who you are, what you already know, what you're stuck on, and what kind of help you want. AI used all of that. There's a name for this skill: prompting. The shortcut: tell AI "I'm [age/grade]. I'm working on [topic]. I'm stuck on [the part]. Can you [the kind of help]?"

"Wait — but how do I know AI's answer is even right?" Great question. AI can get things wrong. Here's the trick: some kinds of answers are more trustworthy than others.

Usually trustworthy: well-known explanations like why volcanoes erupt, how plants grow, how chess pieces move. This stuff is in thousands of textbooks AI learned from. AI almost always gets the big picture right.

Usually NOT trustworthy: specific numbers, exact dates, who-said-what quotes, recent news, and stuff hardly anyone writes about. AI makes these up all the time — and sounds so sure of itself.

How to verify any AI answer: Ask AI: "Can you give me a source where I could check this?" Then go check it. Sometimes AI makes up sources too — so if the book doesn't exist or the article doesn't say what AI claimed, you've caught it. Best rule: AI is a great starting point, not a final answer.

For your grown-up Both prompt/response pairs are real captures from popular AI assistants. The "vague vs specific" gap is the single biggest lever in getting useful AI output, and it's a transferable skill — the same clarity helps with teachers, coaches, and (eventually) co-workers. The pattern shown here (role + context + ask) is a simplified version of frameworks taught in adult AI training (GCES, RTF, etc.) tuned to school-age examples.

On reliability: well-established conceptual explanations (volcano mechanics, photosynthesis, basic chess strategy) are heavily represented in training data and are generally accurate. The danger zones are: specific numbers and dates, recent events, exact quotes, citations and sources, and obscure topics — where citation hallucination is common. The most actionable habit for kids is "ask for a source, then verify the source." This catches both factual errors and the more insidious problem of AI inventing plausible-sounding references that don't exist.

AI4K12 Big Idea 4: Natural Interaction. Effective communication with AI is a learnable skill — the same clarity that helps humans helps AI.

Concept 2 of 4

AI will be an answer-machine OR a tutor — depending on how you ask.

Why this matters: Ask for the answer and you learn nothing; ask it to help you think and you get smarter. Same tool, opposite results.

Exhibit 02

AI as answer-machine vs. AI as tutor.

Same math problem. Two ways to ask. One gives you the answer (and you learn nothing). The other helps you figure it out (and your brain gets stronger). Watch them side by side.

The problem

Loading…

Answer-Machine Prompt

—

AI just hands it over

—

Tutor Prompt

—

AI helps you think

—

What just happened

AI does whatever you tell it to do.

One prompt asked for an answer. The other asked for help thinking. Both got what they asked for — the kid who asked for the answer got it, and learned nothing. The kid who asked for the tutor got smarter.

"Wait — but you just said AI gets things wrong!" Good catch. AI can still get math wrong, even in tutor mode. So why is tutor mode actually safer? Because YOU're checking each step. When AI walks you through a problem, you verify as you go — does this step make sense? Did the numbers add up? If a step is off, you'll catch it. In answer mode, you just get a number with nothing to check against.

The magic words: "Don't tell me the answer — help me figure it out." Try those words in any chatbot.

Pro tip — some AI is built for this: Some chatbots are made specifically for tutoring (Khanmigo from Khan Academy and Socratic from Google are two — your school might use one) and they're tuned to ask questions instead of giving answers. (And there's a clever trick for getting the math itself exactly right — that's the very next activity.)

For your grown-up Both responses are real captures from popular chatbots. Important pedagogical nuance: AI can still get math wrong in tutor mode — the reason tutor mode is safer for learning isn't that AI is suddenly more reliable, it's that the student is doing the verifying step-by-step rather than accepting an answer wholesale. If the AI's walkthrough is wrong, an engaged student is much more likely to catch it.

Some AI is purpose-built for tutoring (Khanmigo, Socratic, MagicSchool) — tuned to ask questions and resist giving answers. Reasoning-focused models (Anthropic's Claude, OpenAI's o-series, Gemini's "thinking" modes) handle math more reliably than general chatbots. And if a chatbot has tool use / code execution enabled, "use a calculator to verify" gets a deterministic answer instead of a token-by-token guess. The phrase "don't tell me the answer, help me figure it out" still reliably activates Socratic-tutor behavior in any modern LLM.

AI4K12 Big Idea 5: Societal Impact. How a person uses AI shapes whether it builds skills or replaces them — same tool, opposite outcomes.

Concept 3 of 4

AI is a guesser — but it can use real tools, like a calculator, to get facts exactly right.

Why this matters: Tools fix the facts, not the understanding. Knowing that lets you ask it to “use a calculator” or “look it up,” then check the parts that matter.

Exhibit 03

AI's secret helpers: tools.

You know AI is a guesser. So how does it ever get hard math exactly right? Here's the clever trick the makers came up with — and the one kind of question no tool can fix.

A tricky one

What is 4,829 × 7,316?

Why this matters

Tools fix the facts. They don't fix the guessing.

On its own, AI guessed at the big multiplication — and got it wrong, but sounded sure. So the people who build AI gave it tools: a real calculator, web search, a way to run code. When AI uses a calculator, the math comes out exact — because a calculator isn't guessing.

But look at the second question. There's no tool for "why is my friend mad." AI just guesses — one general answer for everyone. A calculator can't help, because that answer lives in your real life, not in any data.

Here's the upgrade to your AI brain-map: AI is still a guesser underneath. Tools make its facts trustworthy — math, dates, looking things up — but they can't make it understand. Once you know that, you can get much better answers: when it really matters, ask it to "use a calculator" or "look it up," then check the important parts yourself.

For your grown-up This is the single most useful mental-model upgrade for modern AI. A language model only produces text — a best-guess next token — so it fumbles something like 4,829 × 7,316 done "in its head." But a model with tool use can call a calculator, run code, or search the web, and those tool results are deterministic, not guessed. The same model returns the exact product once it actually invokes a calculator. The durable lesson for kids: tools fix factual accuracy; they don't grant understanding. "Ask it to use a tool, then verify" is the habit that travels.

AI4K12 Big Idea 2: Representation & Reasoning. AI can reason using external tools — a calculator or search engine extends what the model alone can do.

Concept 4 of 4

Three little questions catch most of AI’s mistakes.

Why this matters: “Did you make this up? What if I’m wrong? How could I check?” — carry these and you’re ahead of most grown-ups using AI.

Exhibit 04

Three super-questions for any AI.

You've learned why AI gets things wrong, and how to use AI well. Now: a tiny pocket-sized habit. Three questions to ask any time AI tells you something. Memorize these three and you'll be ahead of most adults.

Memorize these

The Three Super-Questions

1

"Did you make this up?" Catches hallucinations
2

"What if I'm wrong about this?" Beats sycophancy
3

"How could I check this?" Builds the verify habit

Try one — pick a fake AI claim:

Which super-question would you ask?

(pick a claim above)

What just happened

You just used a tool that protects your brain.

Most adults take AI's word for it. You won't, because you have three pocket-sized questions ready. "Did you make this up?" for facts. "What if I'm wrong?" when AI agrees too easily. "How could I check this?" always. In Room 7 you can save these as a printable card for your wall or your school notebook.

For your grown-up These three questions map directly to the three biggest LLM failure modes: hallucination, sycophancy, and unverified claims. They're chosen to be simple enough for an 8-year-old to remember and broad enough that older kids and adults find them useful. Drilling them as a habit (rather than just understanding them) is what turns AI literacy into AI skill.

AI4K12 Big Idea 5: Societal Impact. Critical evaluation of AI output is a learnable, transferable skill — and it starts with simple questions.

End of Room 05

Now you know how to use AI well.

Tell AI about you. Ask for a tutor, not an answer. Carry the three super-questions. That's most of what makes the difference between using AI to skip thinking and using AI to get smarter.

← Room 4 All Rooms Room 6 →