No jargon. No math. Just the plain truth about the thing everyone keeps talking about.
AI for DummiesModule 1No prerequisites
By the end of this lesson you will: know what AI actually is under the hood, understand the nested-dolls hierarchy (AI → ML → DL → LLM), see how a language model predicts its next word, and know the five terms that unlock every AI conversation.
Lesson 01 · Slide 2
The honest definition
Artificial Intelligence (AI) is software that does things we normally associate with human thinking.
What that includes:
Understanding and generating language
Recognising what's in a photo
Answering questions and writing
Planning, predicting, translating
The important part: it's a program. A very sophisticated one — but a program. There's no magic, no brain, no ghost in the machine. The reason this matters: programs can be wrong, and knowing how they go wrong lets you use them safely.
Lesson 01 · Slide 3
AI is not new — what changed?
The term "artificial intelligence" dates to 1956. The chess program that beat Garry Kasparov is from 1997. So why does AI feel like it arrived last Tuesday?
Old AI (rules-based)
A human expert sat down and wrote every single rule:
if "FREE MONEY" in email:
mark_as_spam()
elif "WINNER" in email:
mark_as_spam()
# ... 10,000 more rules
Fragile. Misses anything the author didn't think of.
Modern AI (learning from data)
Instead of rules, show it examples:
# No rules written — just data
model.train(
emails=10_000_000,
labels=["spam", "not spam"]
)
# Model figures out the patterns
The software learns patterns. It can handle email types the engineer never saw.
The current wave is almost entirely learning from data — and the amounts of data are staggering.
Lesson 01 · Slide 4
AI you already use every day
You're not a newcomer to AI — you've been using it for years without the label:
The models behind Claude or ChatGPT are bigger, more general versions of the same principle. The scale is new — the basic idea (learn from examples) has been around for decades.
Lesson 01 · Slide 5
The nested dolls — 90% of the confusion, cleared up
People use "AI," "machine learning," "deep learning," and "LLM" as synonyms. They're not. They nest inside each other:
When someone says "AI" today, they almost always mean the innermost box. But now you know the full picture.
Lesson 01 · Slide 6
How an LLM works — the next-word predictor
At its core, a language model is a next-word predictor. It was trained by reading a huge slice of the internet and books, learning which words follow which. When you type a question, it predicts a likely continuation one word-chunk at a time.
You type: "The capital of France is"
The model now computes probabilities for every possible next word:
Paris
81%
Lyon
11%
known
8%
It picks "Paris," appends it, then predicts the next word given "...France is Paris" — and keeps going until it reaches an end.
This is literally all that's happening. One word at a time. Repeated hundreds of times per response. The model never "thinks" the whole answer at once.
Lesson 01 · Slide 7
Try it: next-word predictions
Type the beginning of a sentence below. This widget uses a small hand-coded probability table — the same idea an LLM uses, just with 30 word patterns instead of hundreds of billions.
A real LLM does this with 70+ billion numbers (its weights) learned from trillions of words. But the fundamental operation — "given what came before, what word fits best?" — is identical.
Lesson 01 · Slide 8
Just bigger autocomplete — really?
Yes. And also: an extraordinary amount of capability emerges from doing that extremely well at enormous scale.
Phone autocomplete
Learned from your texts. Predicts the next 1 word. Knows "I'll be there" → "soon."
Useful for one sentence. Falls apart immediately after.
Training data: ~thousands of your messages.
Claude / ChatGPT
Learned from billions of web pages, books, code, papers. Predicts one word at a time but holds the full context.
Can write an essay, explain a law, debug your code, and then answer a follow-up — coherently.
Training data: trillions of words.
The gap between "phone autocomplete" and "Claude" is not a different technology. It's a different scale of the same technology.
Lesson 01 · Slide 9
What AI is NOT — #1: Not conscious
When an AI says "I think…" or "I feel excited about this!" it is not reporting an inner experience. It's producing text that fits the pattern of how those phrases are used.
There are no feelings, no goals, no self-awareness, no opinions. There is math producing text — sophisticated math, but math.
Why it matters: people over-trust AI that "sounds passionate" or "seems confident." Those signals don't mean anything. A model can sound thrilled about a completely wrong answer.
Lesson 01 · Slide 10
What AI is NOT — #2: Not a search engine
Search engine (Google, Bing)
Indexes documents that actually exist on the web.
Returns links to real sources.
Knowledge updates continuously as the web changes.
Factual errors usually come from bad sources, not the engine itself.
LLM (Claude, ChatGPT)
Generates an answer from patterns learned during training.
Does not look anything up by default — it predicts.
Knowledge has a training cutoff date.
Factual errors come from the model itself — no source to blame.
Some AI tools can search the web as an added capability. But the base model is always a predictor, not a browser. When in doubt, ask: "Did it cite a real URL?"
Lesson 01 · Slide 11
What AI is NOT — #3: Not always right
Hallucination: when an AI states something false as if it were completely true — confidently, fluently, with no hesitation or caveat.
The terrifying part: the tone is identical whether the AI is right or wrong. There is no "hallucination voice." A model sounds exactly as assured when it invents a fake book title as when it correctly names a real capital city.
When AI is reliable
Writing drafts and summaries
Explaining concepts and analogies
Brainstorming and outlining
Well-documented programming tasks
When AI is risky (verify these)
Specific facts, dates, statistics
Book titles, citations, papers
Recent events (past training cutoff)
Medical / legal / financial specifics
Lesson 01 · Slide 12
Hallucination demo — spot the difference
Two AI answers. Both written in the same confident, fluent tone. One is completely true. One is entirely fabricated. Read them, then click to reveal.
Truthful answer
Q: "What's the capital of Australia?"
The capital of Australia is Canberra. It was purpose-built as the capital city and is located in the Australian Capital Territory, roughly halfway between Sydney and Melbourne. Construction began in 1913.
Hallucinated answer
Q: "Tell me about the 2019 novel 'The Bear's Nightmare' by Maria Chen."
"The Bear's Nightmare"
is a
Maria Chen
novel published in
2019
by
Meridian House Press.
The story follows a wildlife biologist who discovers bear attacks near a remote Alaskan station are linked to an
experimental pharmaceutical trial.
Chen's
Booker Prize longlisting
for this novel marked her as a major new voice in ecological thriller writing.
Both are written identically. That is the danger.
Lesson 01 · Slide 13
The mental model that fixes everything
Treat AI like a very fast, very well-read intern. Useful — sometimes brilliant — but check anything that matters against a real source.
Great intern qualities:
Reads everything you give them in seconds
Writes excellent first drafts
Explains complex things simply
Never bored by repetitive tasks
Instantly switches topics
Intern limitations:
Can sound confident while being wrong
May invent citations to impress you
Knowledge has an expiry date
No common sense about your context
Not responsible for their mistakes — you are
You don't need to understand how an engine works to drive a car. You just need to know that engines can stall. Same here.
Lesson 01 · Slide 14
Tokens — what the model actually reads
AI doesn't read words. It reads tokens — chunks of text roughly ¾ of a word on average. This matters for two practical reasons: billing and memory.
The sentence "Unhelpfulness is never trivially safe." splits into 7 tokens:
Unhelpfulness is never trivially safe.
Common words = 1 token. Unusual words get split into sub-word pieces. Punctuation is its own token.
Context window: the maximum tokens the model can hold in memory at once. Think of it as a desk that fits a certain number of pages. Fill it up, and the model "forgets" the beginning.
Billing: cloud AI services charge by the token — both what you send (input) and what comes back (output). ~750 words ≈ 1,000 tokens.
Lesson 01 · Slide 15
The 5 terms that unlock every AI conversation
Learn these and you'll be able to follow — and participate in — almost any discussion about AI tools, pricing, or limitations.
Term
Plain meaning
Real example
Model
The trained AI itself — like a swappable engine.
Claude Opus 4, GPT-4o, Llama 3.3
Prompt
What you type to the AI — your question or instruction.
"Summarise this in 3 bullet points"
Token
A chunk of text (~¾ word). AI reads and bills in tokens.
1,000 tokens ≈ 750 words
Context window
How much the model can "hold in mind" at once.
200 K tokens ≈ ~150,000 words
Hallucination
When the AI states something false as true.
Inventing a book title or citation
These five terms appear in every AI product page, pricing table, and support ticket. Knowing them fluently makes you a more effective user immediately.
Lesson 01 · Slide 16
What AI is good at — and what it isn't
The habit that matters most in this course: use AI freely for drafts, explanations, and brainstorming. Verify anything with a real-world consequence — facts, numbers, citations, legal or medical specifics — against a trusted source.
Key insight: AI is not all-or-nothing. Being a good AI user means knowing where on this spectrum your current task sits.
Lesson 01 · Slide 17
Mental model: four sentences
It's a predictor, not a knower. "Likely" is not the same as "true."
Confidence is not a signal. AI sounds equally sure whether right or wrong.
It's enormously useful anyway. First drafts, explanations, brainstorming — a fast predictor is genuinely powerful.
Verify anything that matters. This habit runs in the background in every lesson of this course.
Lesson 01 · Slide 18
Recap & what's coming
What you now know:
AI = software that mimics thinking — it's a program
Modern AI learns from data, not rules
AI ⊃ ML ⊃ Deep Learning ⊃ LLMs (nested dolls)
LLMs are next-word predictors at huge scale
Hallucinations are real, undetectable by tone
The 5 terms: Model, Prompt, Token, Context window, Hallucination
Coming up:
Lesson 02 — the types of AI that exist (narrow, generative, multimodal)
Lesson 03 — the four ways to use AI (Chat, API, CLI, Code tools)
Lesson 04 — installing Claude, LM Studio, Ollama
Lesson 05 — how to prompt well — the skill that changes everything
Up next: Read the Theory page for three interactive demos — the nested-dolls animation, the next-word predictor, and the hallucination toggle. Then do the Homework.