What AI Is—and Is Not
What AI Is—and Is Not
AI & Prompt Engineering for Everyone · double-click any text to add a note · hover dotted terms for definitions
Tutorial 1 of 17
Absolute Beginner
No-code
1 Course position and tonight’s promise
Day 1 begins after your Day 0 readiness pack. Tonight you will sort everyday examples into old software, predictive AI, or generative AI, then keep a named “AI Tool Choice Card” with a reusable prompt and verification checklist.
Promise: in 60 minutes you can explain what AI does, what it does not guarantee, and choose a safer tool for a real non-sensitive task.
2 What you will learn and why it matters
- Spot AI already present in familiar apps.
- Contrast fixed rules with learned patterns.
- Explain generative AI as next-piece prediction without treating it as a person.
- Use prompt, model, context, hallucination, and credits accurately.
- Verify output, protect sensitive data, and choose a fitting tool.
These decisions prevent wasted credits, privacy exposure, and confident-looking errors before later portfolio projects become more complex.
3 A familiar situation: predict before you choose
Fictional sample values: Meena’s Corner Shop advertises ₹800 notebooks with a fixed 10% festival discount. A calculator can apply the rule; a predictive system might estimate demand from past totals; a generative tool might draft a short poster.
| Task | Your prediction | Best starting mechanism |
|---|---|---|
| Compute ₹800 minus 10% | Same answer every time | Fixed rule/calculator |
| Estimate next week’s demand | Varies with learned historical patterns | Predictive AI |
| Draft three poster lines | New wording from the instruction and context | Generative AI |
Prediction question: which result must be exact, which is an estimate, and which is a draft? Write your answer before continuing.
4 Simple explanation, then accurate vocabulary
AI is software that performs tasks associated with human intelligence, often by finding patterns in data. It is not a person, intention, consciousness, or automatic source of truth.
| Term | Plain meaning | Accurate meaning |
|---|---|---|
| Rule-based software | Follows written instructions | Executes human-authored conditions; it does not learn the rule from examples. |
| Predictive AI | Estimates what is likely | A model maps inputs to a prediction using patterns learned during training. |
| Generative AI | Creates a new draft | A model predicts likely next pieces—such as text tokens or image elements—conditioned on its input. |
Difficult idea: “AI understands” is convenient shorthand, not proof of human-like understanding. தமிழ்: AI மனிதரைப் போல புரிந்துகொள்கிறது என்று கருத வேண்டாம்; அதன் பதிலைச் சரிபார்க்கவும். Keep the English term “model.”
5 AI already present in everyday apps
AI can appear quietly in spam filtering, photo grouping, speech-to-text, recommendations, translation, fraud alerts, and route estimates. Each feature has a narrow job and can be wrong.
| Everyday feature | Likely role | Human check |
|---|---|---|
| Email spam filter | Predictive classification | Inspect spam before deleting important mail. |
| Keyboard suggestions | Next-piece prediction | Read before sending. |
| Photo search by object | Learned visual matching | Confirm the people or objects yourself. |
| Alarm at 07:00 | Usually a fixed rule | Check time, repeat, and time zone. |
A feature label alone cannot prove its implementation. Ask what input it uses, what output it produces, and how errors are corrected.
6 Rule-based software compared with learned patterns
Primary example: the shop’s fixed discount is “if eligible, multiply price by 0.90.” A learned model instead studies fictional past shopping rows and estimates whether a customer might use an offer. The model produces a probability, not the exact discount rule.
| Evidence | Fixed shopping rule | Learned shopping pattern |
|---|---|---|
| Instruction | Written directly by a person | Parameters fitted from examples |
| Same eligible ₹800 input | ₹720 every valid run | A score that may change with model/data |
| Best use | Exact, auditable calculation | Uncertain pattern estimate |
| Failure | Wrong or incomplete rule | Biased, stale, unrepresentative data |
Before evidence: ₹800. After applying the stated 10% sample rule: ₹720. Independently check 800 × 0.90 = 720; do not ask a language model to replace the calculator.
7 Generative AI and next-piece prediction
A generative AI model creates output one likely piece at a time, influenced by its training and the current input. In text, a piece may be a token rather than a whole word. This can produce useful new wording without retrieving a stored, verified answer.
| Prompt start | Plausible continuation | What it proves |
|---|---|---|
| “Write a five-word shop poster…” | “Save ten percent on notebooks.” | The constraint influenced likely wording. |
| “Name our invented 1842 award…” | A convincing invented name | Fluency does not prove the award exists. |
Simplification: the lesson shows a straight sequence. Real models use numerical representations, many layers, probabilities, system instructions, and sampling; the diagram does not expose private reasoning.
8 Prompt, model, context, hallucination, and credits
A prompt is the instruction and input you send. A model is the trained pattern system that transforms that input. Context is the information available for this response, including relevant instructions and supplied material.
A hallucination is plausible-looking unsupported or false output. Credits are provider-defined usage units or allowances; they may measure messages, tokens, media generations, compute, or something else.
| Object in this lesson | Technical connection |
|---|---|
| Input card | The actual prompt plus safe sample data |
| Tool-operation card | The chosen fixed rule, predictive model, or generative model |
| Source card | Context/evidence supplied for checking |
| Human review gate | A person compares claims and format with evidence |
| Saved choice card | The exported artifact—not proof that every claim is true |
Check current limits, credit rules, watermarking, retention, and commercial rights only in the provider’s official terms; these can change.
9 Five beginner myths
| Myth | Reality | Better action |
|---|---|---|
| “AI is one magic tool.” | Different systems do different jobs. | Match the mechanism to the outcome. |
| “It searches the internet automatically.” | Access to current sources varies. | Confirm sources and access explicitly. |
| “Confident means correct.” | Fluency can accompany invention. | Check each factual claim. |
| “More personal data improves everything.” | Extra data adds privacy risk. | Minimise or use fictional samples. |
| “The first draft is final.” | Generation is an editable draft. | Review and revise from evidence. |
Do not use “AI” as a blanket explanation. Name the input, operation, output, risk, and human check.
10 Verification and sensitive-data safety
Verify facts, links, extracted data, calculations, names, dates, and answer keys against an authoritative source or manual check. When a claim has no dependable source, label uncertainty or remove it—do not repeatedly retry and hope.
- Never upload Aadhaar or PAN numbers, banking credentials, passwords, medical records, employer-confidential material, or unnecessary personal data.
- Use the fictional shop values or a no-upload paper exercise.
- AI explanations are not legal, medical, or financial advice; seek a qualified professional for high-stakes decisions.
- For children, AI explains and the child writes; an adult or teacher reviews.
Blocked state: if input is sensitive or a claim is unsupported, stop at that exact stage. Remove/replace the data or locate evidence before proceeding.
11 Choosing the right AI tool
| Need | Best starting tool | Reason / check |
|---|---|---|
| Exact known discount | Calculator or spreadsheet rule | Auditable arithmetic; verify formula. |
| Predict demand from suitable records | Predictive system with evaluated data | Designed for estimates; inspect error and bias. |
| Draft poster wording | Generative text tool | Designed to create; fact-check every claim. |
| Find a current official policy | Official site/search with source links | Currency and provenance matter. |
| High-stakes diagnosis or advice | Qualified professional | AI output is not professional advice. |
- Define the outcome and audience.
- Decide whether it needs a fixed rule, a prediction, generation, retrieval, or a professional.
- Check acceptable data, source access, export format, accessibility, credits, watermark, rights, and retention.
- Run a small fictional test; verify; then keep or reject the tool.
Tool names are replaceable examples. Interfaces and plan terms change; use the visible control labelled New chat/Create/Export or its current equivalent, and verify official help.
12 Interactive process model: route, review, revise
First predict a route. Then choose “Try a choice” and compare the visible outcome. States progress from Not started → Working → Needs review → Verified → Ready; sensitive input or unsupported claims become Blocked and return to the failed stage.
🎬 Everyday input to a verified result
The sample artifact moves conceptually through these cards. Use all controls; reduced-motion mode changes state without travel effects.
Input (prompt)
Not started → working
➜
Choose operation
rule / predictive / generative
➜
Prediction or creation
needs review
➜
Human verification gate
source + safety
➜
Saved artifact
verified → ready
| Stage | Accepted example | Rejected / correction |
|---|---|---|
| Input | Fictional ₹800 price | Bank details → replace before upload |
| Operation | 10% fixed rule for arithmetic | Generative tool for exact arithmetic → use calculator |
| Output | “10% off notebooks” | “Award-winning since 1842” → remove without evidence |
| Review | 800 × 0.90 = 720 | No source → blocked |
| Export | Named choice card + checklist | Chat only → save and reopen |
Textual fallback: predict; submit safe sample; route it; inspect one change; check facts and privacy; return a failure to the exact stage; export only after it passes.
13 Facilitator-led demonstration: poor input to verified result
Poor starting input: “Make a great discount ad.” It lacks audience, values, format, source limits, and privacy boundaries. The first fictional output says “Award-winning since 1842”—a realistic hallucination because the prompt invited unsupported decoration.
Revised reusable prompt (copy only after diagnosing the failure)
Role: Help a beginner shop owner draft wording.
Audience: Local notebook buyers.
Use only these fictional facts: price ₹800; discount 10%; final price ₹720.
Output: one headline and one sentence, each under 12 words.
Limits: Do not invent awards, history, urgency, stock, or customer claims.
After the draft, list each factual claim for human verification.| Visible decision | Before | After / evidence |
|---|---|---|
| Scope | “great ad” | Audience, output, and limits stated |
| Arithmetic | Unspecified | ₹720 checked with 800 × 0.90 |
| Invented claim | “Award-winning since 1842” | Removed because no supplied source supports it |
| Result | Polished but unsafe | Constrained draft plus claim list |
Narrate visible evidence and decisions, not private chain-of-thought. GUI path: open the current text input, paste only fictional data, submit, inspect, revise, then use Export/Copy/Save if available. Interfaces change.
14 The AI Tool Choice Card
Portfolio template
Artifact name: Day-1-AI-Tool-Choice-Card-[your initials]
Safe task and audience:
Required output and limits:
Mechanism: fixed rule / predictive / generative / retrieval / professional
Why this mechanism fits:
Non-sensitive context/source:
Prediction before run:
Result and verification evidence:
Revision and reason:
Tool limits/credits/rights checked at: [official page + date]This card is both the finished artifact and an explanation of the decision. Keep the reusable prompt and checklist with it so another person can reproduce the safe method.
15 Learner lab: build your safe Tool Choice Card
- Prepare fictional or minimised input; use the shop sample if you prefer no upload.
- Write your predicted output and one likely risk.
- Choose fixed rule, predictive AI, generative AI, retrieval, or a qualified professional—and explain why.
- Run the first attempt using current GUI controls or complete it on paper.
- Check requested format, each fact, calculation, link, and privacy boundary against evidence.
- Revise the exact failed instruction or context; never use blind retries instead of a source.
- Export/save the named card, prompt, output, and checklist; reopen the file.
- Add: “I revised ___ because the evidence showed ___.”
Completion checkpoints: no sensitive data; prompt names audience/output/limits; format matches; factual details are independently checked; you can explain one revision.
| Checkpoint | Evidence to show |
|---|---|
| Safe input | Fictional label or redacted/minimised source |
| Clear request | Audience + output + limits |
| Format | Manual comparison with request |
| Accuracy | Source or calculation beside every factual claim |
| Revision | Before/after plus one-line reason |
| Kept work | Named file reopens correctly |
16 Behind the scenes, limitations, safety, and verification
- The prompt card represents submitted instructions and safe context; hidden provider instructions may also affect output.
- The model/tool card represents mathematical operations, not a mind or truth engine.
- The source card represents evidence a human can inspect; model output is not its own source.
- The review gate represents privacy, accuracy, format, permissions, and terms checks.
- The exported artifact represents what you saved, not a guarantee or professional approval.
Models may reflect training gaps or bias, lose important context, produce inconsistent outputs, and lack current information. Verify facts, links, extracted data, and answer keys. Confirm current credits, limits, watermarks, retention, and commercial rights from official terms.
17 Common confusion, troubleshooting, and best practices
| Symptom | Likely cause | Safe fix |
|---|---|---|
| Fluent output is wrong | Tone mistaken for evidence | Stop and compare claims with an authoritative source. |
| Generic output | Too little context | Add audience, desired output, safe facts, and limits. |
| First draft is “almost” right | No review/revision checkpoint | Mark the precise mismatch and revise from evidence. |
| Private data appears | Sensitive input was submitted | Stop sharing; follow provider/incident guidance and replace it—do not retry. |
| Feature unavailable or marked | Plan/credit/watermark limit | Keep required marks or choose a permitted alternative; check official terms. |
| Work disappears | Relied on chat history/autosave | Export named files and reopen them. |
| Repeated answers disagree | No reliable source or unsuitable tool | Stop retries; find a source, use a rule/calculator, or ask a qualified person. |
| Exact discount varies | Generative tool chosen for arithmetic | Use the stated rule in a calculator/spreadsheet and verify. |
- Best practice: define the job before the brand.
- Best practice: minimise data and separate facts from draft language.
- Best practice: preserve before/after evidence.
- Trainer tip: ask “Which visible evidence changed your decision?”
18 Takeaways
- AI is a collection of task-specific systems, not a person or truth guarantee.
- Use explicit rules for declared calculations, predictive systems for evaluated estimates, and generative systems for drafts.
- A prompt supplies instructions; context supplies available information; the model performs the operation.
- Fluent output can hallucinate. Verify independently and block sensitive data.
- Keep the Tool Choice Card, reusable prompt, checklist, evidence, revision, and reflection.
Your Day 1 portfolio artifact is complete when you can reopen it and explain why the mechanism fit, which evidence verified it, and what you revised.
19 What comes next
Next is Topic 02—Build Strong Prompts. Open its page when you are ready, and keep today’s safe Tool Choice Card: the next lesson will strengthen its audience, output, context, and limits.
- Reopen your named artifact and reusable prompt.
- Keep the fictional shop sample available.
- Write one question about a weak prompt for the next session.
18 Knowledge check—submit before feedback
Pick an answer for each question, then press Check answer. (Notes are disabled in this tab.)
1. Which statement best defines AI?
✅ Correct!
❌ Not quite — the correct answer is .
AI is software performing particular pattern-related tasks; consciousness, ordinary fixed rules, and guaranteed truth do not follow. Review #key-concepts.
2. Which mechanism best computes a declared 10% discount on ₹800?
✅ Correct!
❌ Not quite — the correct answer is .
The declared arithmetic is exact and auditable with 800 × 0.90 = 720. Other choices add uncertainty. Review #rule-based-software-compared-with-learned-patterns.
3. Predict what should happen when a poster draft invents an 1842 award.
✅ Correct!
❌ Not quite — the correct answer is .
An unsupported claim fails the human verification gate; fluency cannot supply evidence. Review #interactive-process.
4. Which input is safe for the first practice?
✅ Correct!
❌ Not quite — the correct answer is .
Fictional values avoid unnecessary exposure. The other inputs are sensitive or confidential. Review #verification-and-sensitive-data-safety.
5. Results keep disagreeing and there is no cited source. What next?
✅ Correct!
❌ Not quite — the correct answer is .
Disagreement signals a verification or tool-choice problem. More guesses and private data do not create evidence. Review #best-practices.
19 Exercises—attempt before reveal
Try answering each question yourself before expanding the model answer.
1. Modify: Change the shop poster audience to school administrators or change its word limit. Record what changed. Reveal after attempting.
A strong response identifies a visible vocabulary, tone, or length change while retaining the verified fictional facts and safety limits.
2. Extend: Add a second fictional input or a five-line verification checklist. Reveal after attempting.
Check privacy, required format, every factual claim, calculations/links, and current official tool terms; record the evidence.
3. Apply: Use the choice method on one real, non-sensitive personal task. Reveal after attempting.
Define the need, choose the mechanism, predict, use minimised input, verify independently, revise from evidence, export, and reflect.
4. Optional: Teach the workflow to someone and capture their question—not their personal data. Reveal after attempting.
Record a general question and point to the workflow stage that resolves it; omit names and identifying details.
20 Quick-review cards
Click a card to reveal the back.
Rule or learned pattern?
A rule is explicitly written; a learned model fits patterns from examples.
Generative AI
A model that creates new output by predicting likely next pieces conditioned on input.
Hallucination
Plausible-looking output that is false or unsupported; check independently.
Right-tool test
Match the required outcome to rule, prediction, generation, retrieval, or professional help.
Ready state
Safe input, suitable method, verified result, explained revision, and reopened export.
21 Facilitator and interview Q&A
1. Explain AI simply without calling it a person.
AI is software that performs particular tasks by using rules or patterns; its useful output can still be wrong and needs human checking.
2. Demonstrate the shopping workflow.
Show the fictional ₹800 input, route exact 10% arithmetic to a calculator, route poster wording to generation, verify ₹720 and every claim, revise, then reopen the saved card.
3. Diagnose a fluent but weak poster.
Point to missing audience/limits and the unsupported award claim; compare with supplied facts, remove the claim, and show before/after evidence.
4. Identify a privacy and accuracy risk.
A customer account number exposes sensitive data; an invented award is inaccurate. Block both at their stages rather than publishing or retrying.
5. Adapt the explanation for a child or senior learner.
Use the alarm-versus-keyboard example, short labels, fictional data, and hands-on checks; for children, AI explains and the child writes under adult review.
22 Glossary
- AI
- Simple: software that does tasks that can seem intelligent. Technical: computational systems performing tasks such as learned pattern inference. Page: /Tutorials/ai-for-everyone-01-what-ai-is-and-is-not
- generative AI
- Simple: AI that creates a new draft. Technical: a model generating likely next pieces conditioned on input. Page: /Tutorials/ai-for-everyone-01-what-ai-is-and-is-not
- model
- Simple: the pattern system a tool uses. Technical: a parameterised mathematical system fitted or configured to map inputs to outputs. Page: /Tutorials/ai-for-everyone-01-what-ai-is-and-is-not
- prompt
- Simple: what you ask or provide. Technical: instructions and input supplied to condition a model response. Page: /Tutorials/ai-for-everyone-01-what-ai-is-and-is-not
- context
- Simple: information available for this response. Technical: instructions and content within the model’s usable input window. Page: /Tutorials/ai-for-everyone-01-what-ai-is-and-is-not
- hallucination
- Simple: a believable-looking invention or mistake. Technical: generated content that is false or unsupported by reliable evidence. Page: /Tutorials/ai-for-everyone-01-what-ai-is-and-is-not
- credits
- Simple: a provider’s usage allowance or units. Technical: provider-defined metering units that may govern messages, tokens, media, or compute. Page: /Tutorials/ai-for-everyone-01-what-ai-is-and-is-not