M2-B — How AI Generates Answers
The core idea When you ask ChatGPT something, it doesn't open a textbook. It does one thing on loop: read your words → predict next word → repeat.
⌨️ What Happens When You Type a Prompt
Your prompt is really giving AI four things at once:
| What you're giving | What it means |
|---|---|
| Situation | The context around your question |
| Task | What you actually want |
| Style (output format) | How you want it to sound |
| Limits | What to stay away from |
The steering wheel Your prompt is the steering wheel. The model is the engine. Without steering, it goes anywhere.
🔄 Step-by-Step — How an Answer is Built
You type a prompt
↓
System breaks your text into small pieces (tokens)
↓
Model looks at tokens and asks: "What's the best next token?"
↓
It picks one (with a bit of randomness)
↓
Adds it to the answer
↓
Repeats until the answer feels complete
What's a token? Just a small chunk of text — a word or part of a word. You don't need to memorise the word. Just know: small pieces.
🎬 A Tiny Demo
Prompt: "In my school, the best teacher is"
The model predicts step by step: the → one → who → explains → with → examples
Result: "In my school, the best teacher is the one who explains with examples…"
This is it. It completes text smartly. That's the whole trick.
🎲 Why the Same Prompt Gives Different Answers
The model doesn't pick the next word in one fixed way. It picks from a group of "good next words" — like a weighted dice.
| Setting | What it does | What you notice |
|---|---|---|
| More strict | Picks the most common next word | Safe, consistent, repetitive style |
| More creative | Allows less-common words | More variety, occasionally more mistakes |
The real insight ChatGPT is not copying a stored answer. It is creating a new answer every time — step by step.
✍️ Prompt Quality = Output Quality
Your prompt gives the model its direction. A vague prompt = AI guesses.
Unclear vs Clear — Side by side
| Unclear prompt | Clear prompt |
|---|---|
| "Explain AI." | "Explain AI to a 10th student using a school example, in 8 bullets, no technical words." |
Think of it like a recipe A clear prompt gives: ingredients + steps + style. A vague prompt gives: the chef makes something random.
📋 Prompt Checklist
Before you send a prompt, ask yourself:
| Part | Question to ask yourself | Example |
|---|---|---|
| Goal | What exactly do I want? | "Make study notes" |
| Audience | For whom? | "10th standard students" |
| Format | How should it look? | "Headings + recap" |
| Limits | What to avoid? | "No jargon, no math" |
| Examples | Real-life context? | "School, cricket, shopping" |
🔍 AI is NOT Google — Clear Difference
| ChatGPT | ||
|---|---|---|
| How it works | Generates text from scratch | Finds pages that already exist |
| Gives you | Explanation, writing, drafts | Links and sources |
| Best for | Understanding and creating | Checking facts and finding official info |
| Risk | Can confidently generate something false | Can give outdated or irrelevant links |
Use both — not just one Use ChatGPT to understand and draft. Use Google to verify facts.
✅ Recap
30-second read
- Your prompt is broken into small pieces (tokens), and the model predicts next piece, again and again.
- Answers are created step-by-step — not fetched from a stored page.
- Same prompt can give slightly different answers because of built-in randomness.
- Better prompts work because they give clearer direction: audience, format, limits, examples.
- ChatGPT explains and drafts well. Google verifies and finds facts.