Ten lessons ago you didn’t know what an LLM was. Today you can install a local model, call an API, and build a working bot. This slide deck is your permanent reference — vocabulary locked in, cheat sheet packed, road ahead mapped.
Bookmark the Theory page — it has a searchable glossary, a printable cheat sheet, and a 4-week checkbox tracker.
Lesson 10 · Slide 2
Questions you can now answer
Before this course, these were mysteries. Now they’re old news.
What’s the difference between AI, ML, deep learning, and an LLM?
Why does AI make things up — and what’s a hallucination?
What is a token, and why does it matter for cost?
What is a context window and what happens when you exceed it?
How do I run a model on my own computer with no data leaving?
What is an API key and how do I keep it safe?
How do I write a Python script that calls an AI?
What is RAG, and when do I need it?
What is an agent — and what makes it different from a chat?
What is AGI — and does it exist yet?
Lesson 10 · Slide 3
Glossary spotlight — core concepts
Core
AI
Software doing tasks that normally need human intelligence — language, images, planning, prediction.
Core
Machine Learning
AI that learns patterns from data instead of hand-written rules.
Core
Deep Learning
ML using multi-layer neural networks. Responsible for most AI progress since 2012.
Core
LLM
Large Language Model: a text-trained model like Claude, ChatGPT, or Gemini. Predicts the next token.
Lesson 10 · Slide 4
Glossary spotlight — working with AI
The four terms that show up in every single AI conversation.
Work
Prompt
What you type to the AI. The whole craft of getting good results is learning to write better prompts. Recipe: Role + Context + Task + Format.
Work
Token
A chunk of text — roughly ¾ of a word. AI reads and charges by the token, not words or characters. “Hello world” ≈ 3 tokens.
Work
Context window
How much text the model can consider at once. Past the limit, the model “forgets” the beginning of your conversation.
Risk
Hallucination
Confident, false AI output stated as fact, with the same fluent tone as correct answers. You cannot tell from tone alone. Always verify what matters.
Remember: confident tone ≠ accurate. The model sounds equally sure whether it is right or completely wrong. Verify anything consequential.
Lesson 10 · Slide 5
Glossary spotlight — doorways & tools
Tool
API
Application Programming Interface — the way your programs send prompts to and get answers from an AI service over the internet.
Tool
CLI / Terminal
The text window where you type commands. Used to run Ollama, Python scripts, and Claude Code. Looks intimidating; it’s just typing.
Tool
Local model
A model running on your own computer. No data leaves your machine. Free, private, offline-capable.
Work
Agent
AI that takes multi-step actions toward a goal — can call tools, browse the web, or run code. Claude Code is an agent.
Work
RAG
Retrieval-Augmented Generation — feeding the AI snippets of your own documents so it answers from them instead of guessing.
Core
Multimodal
A model that handles more than text — also images, audio, video, or documents. Most modern frontier models are multimodal.
Lesson 10 · Slide 6
Glossary spotlight — the bigger picture
Big picture
AGI
Artificial General Intelligence — hypothetical human-level general AI that can do anything a human can. Does not exist yet. Much debated. Don’t confuse it with today’s impressive-but-narrow systems.
Big picture
Parameters
A model’s internal knobs — “7B” means 7 billion parameters. More parameters generally means smarter but heavier and slower.
Big picture
Generative AI
AI that creates new content: text, images, audio, video, or code. LLMs are one type; image generators (Stable Diffusion, DALL·E) are another.
Narrow AI is all that exists today — good at specific tasks. AGI (human-level general capability) is a future hypothesis, not a current product.
Model sizes: 3B = phone-friendly • 7B = laptop • 70B = needs a powerful GPU • >100B = datacenter. Start small, scale up as needed.
Lesson 10 · Slide 7
One-page cheat sheet — part 1
To just use AI
Go to claude.ai and type. That’s it.
Free tier is real — you can do useful work without spending anything.
If the first answer is wrong, say what was wrong and ask again
Be specific — vague question = vague answer
Example prompt using the recipe: “You are a friendly lawyer [role]. I have a dispute with my landlord about a broken boiler [context]. Explain what rights I have under UK law [task]. Use bullet points, plain English, no jargon [format].”
Lesson 10 · Slide 8
One-page cheat sheet — part 2
To run AI privately on your computer
Easy route (GUI): Download LM Studio, search for a model, download, run
Alternative: Cursor (cursor.sh) — VS Code fork with AI built in
Hardware check: 8 GB RAM runs small models; 16 GB is comfortable; 32 GB+ for the big ones. Apple Silicon Macs are excellent for local AI.
Lesson 10 · Slide 9
One-page cheat sheet — part 3
To automate with code
Get an API key at console.anthropic.com
Store it in a .env file: ANTHROPIC_API_KEY=sk-ant-...
Install the SDK: pip install anthropic python-dotenv
Start with a cheap model (claude-haiku-4-5) & set max_tokens
Only upgrade to a larger model once logic is confirmed working
Always (the three rules)
Verify what matters. AI is a confident guesser, not a fact database.
Protect private data. Don’t paste passwords or confidential records into a cloud chat.
Keep a human in the loop for anything medical, legal, financial, or safety-critical.
The full interactive cheat sheet (expandable • printable • 6 sections) lives on the Theory page. Hit Expand all then ⌘P to print a one-pager.
Lesson 10 · Slide 10
Your 4-week path to confident AI use
This is a practice plan, not a course. Each week adds one capability. The Theory page has a full checkbox tracker — state saved in your browser.
Lesson 10 · Slide 11
Week 1 spotlight — Chat only
Week1
Goal: build the habit of actually using AI every day
Use Claude.ai for at least one real task every day (writing, research, summarising)
Write a prompt using the full recipe: Role + Context + Task + Format
Take a mediocre answer and iterate on it until it’s good
Catch the AI making something up — verify with a real source and note the discrepancy
Explain what an LLM is to someone who doesn’t know — without jargon
Don’t move on until: you feel comfortable iterating on an answer with follow-up messages, and you’ve caught at least one hallucination.
Tip: Use it for something you actually care about — a real email, a real question, a real problem at work. Fake exercises don’t build habits.
Lesson 10 · Slide 12
Week 2 spotlight — Go local
Week2
Goal: run AI on your own machine with no internet
Install LM Studio (lmstudio.ai) or Ollama (ollama.com)
Download a small local model (3B–7B parameters)
Ask the same question to both your local model and Claude.ai — compare quality
Turn off WiFi and confirm your local model still works
Quick-start commands (Ollama):
ollama pull llama3.2:3b
ollama run llama3.2:3b
First run downloads the model (~2 GB). After that it loads in seconds.
Lesson 10 · Slide 13
Week 3 spotlight — Touch the API
Week3
Goal: make your first automated AI call from Python
Get an Anthropic API key at console.anthropic.com
Store the key in a .env file — never in your code
Run the Hello Claude Python example from Lesson 6
Process a small batch with a loop (e.g., summarise 5 items)
Check your usage console — understand what you spent
# The whole thing in 8 linesimport anthropic
client = anthropic.Anthropic() # reads ANTHROPIC_API_KEY from env
msg = client.messages.create(
model="claude-haiku-4-5",
max_tokens=256,
messages=[{"role": "user", "content": "Say hello in one sentence."}],
)
print(msg.content[0].text)
Lesson 10 · Slide 14
Week 4 spotlight — Build one thing
Week4
Goal: ship something real, however small
Pick one starter project from Lesson 8 (or write your own idea)
Get a “hello world” working — even if it’s ugly
Make it do the one thing you actually wanted it to do
Show it to one other person (even if they aren’t technical)
Starter project ideas
Summarise all your email newsletters in one daily digest
AI that answers questions about your own notes / documents
Script that rewrites your job descriptions in plain language
CLI tool that translates files in bulk
The mindset shift: You’re not trying to build the perfect thing. You’re trying to go from zero-to-shipped once — because the second project is always easier than the first.
AI is confidently wrong sometimes. Verify what matters.
What “verify” looks like: if the AI tells you a legal deadline, a drug interaction, a financial figure, or any claim you’re going to act on — check it against a primary source. The AI is not that source.
What it does NOT mean: you don’t need to fact-check every sentence. Use judgment. Verify the things that matter; let the draft writing slide.
This habit — more than any technical skill — is what separates a confident, responsible AI user from someone who gets burned. Build it in Week 1 and never lose it.
Lesson 10 · Slide 17
You finished AI for Dummies
You don’t need to understand how an engine works to drive a car. The same is true here. Start with Chat, build the verify-everything habit, and let curiosity pull you toward the next thing. That’s it.