M2-E — Your Final Mental Model of AI
Apr 29, 2026 3:38 PM
What this note is for Module 2 is not about memorising new words. It's about getting one clear picture in your head — how does AI actually behave?
🖼️ The Final Picture — "Smart Autocomplete with a Memory Window"
Here's all you need to hold in your head:
1. AI reads your message
2. It looks at the chat so far (only what fits in its window)
3. It predicts the next word
4. It keeps predicting until the answer looks complete
That's it. Everything else is details around this loop.
The real insight ChatGPT is not "a person who knows." It is "a system that predicts" — and prediction can be brilliant and wrong.
📦 The 3-Box Model — Think Clearly About Any AI Output
Every answer you get is shaped by three boxes:
| Box | What it means | What you control |
|---|---|---|
| Input | Your prompt + chat context | A lot ✅ |
| Model | The trained prediction engine | Not directly ❌ |
| Output | The generated answer | You can review and fix ✅ |
If the output is bad, first improve the input. That's where your power is.
⚖️ Patterns vs Reality — The One-Line Rule
AI is best at patterns, not at reality.
| Patterns (AI is strong) | Reality (AI can fail) |
|---|---|
| Writing style and tone | Today's news or live data |
| Explaining ideas clearly | Exact rules for your school |
| Organising information | Your personal life details |
| Brainstorming examples | Official legal/medical facts |
For anything in the "Reality" column → you must verify.
😎 Why AI Can Sound Right and Still Be Wrong
| Looks right because… | But might be wrong because… |
|---|---|
| Smooth, fluent explanation | Missing a key fact |
| Confident, teacher-like tone | Guessing without a real source |
| Detailed, step-by-step flow | Step 3 is quietly invented |
Nice writing is not proof. Don't use it as one.
🎲 Why the Same Prompt Gives Different Answers
The model picks from multiple "good next words" each time — so small randomness is baked in. Different phrasing, different examples, different order can all happen from one prompt.
How to reduce randomness:
- "Give exactly 6 bullets"
- "Use only school examples"
- "Do not add new claims; only rewrite what I gave you"
More specific = more consistent.
🧑💼 The Mature Way to Use AI
Think of AI like an intelligent junior assistant:
You ask it to:
- explain
- draft
- organise
- brainstorm ideas
You still do:
- verification
- final decision
- take responsibility
✅ Final Checklist — Every Time You Get an AI Answer
Before acting on any AI output, run through this:
High-stakes = don't trust more. Verify more.
🗺️ Module 2 — Full Map at a Glance
| File | What it covered |
|---|---|
| M2-A | What's inside AI — prediction, patterns, not a brain |
| M2-B | How answers are generated — tokens, prompts, randomness |
| M2-C | Why AI makes mistakes — hallucination, confidence, context limit |
| M2-D | How to use AI correctly — habits, workflow, templates |
| M2-E | Final mental model — 3-box framework, checklist ← you are here |
✅ Recap
30-second read
- Final mental model: ChatGPT = smart autocomplete with a limited memory window.
- Answer quality = Input → Model → Output. Improve input first.
- AI is strong at patterns (language), weaker at reality (fresh facts).
- It can sound correct even when wrong — confidence is not proof.
- Safe use = AI drafts + humans verify and decide.