Let's talk about what AI actually is.
The course starts here.
In the next 22 minutes, I'm going to give you the vocabulary that makes everything else in AI make sense. By the end you'll know what an LLM actually is, why your prompts get charged by the token, what hallucinations are, and how to write a prompt that doesn't waste your time.
AI, ML, GenAI, LLMs — the cast list.
You'll hear all four words used like they mean the same thing. They don't.
- AI — any system that simulates human intelligence. Includes spam filters and your Roomba.
- ML (Machine Learning) — subset of AI that learns from data. Inventory forecasts, dog-breed classifiers, self-driving cars.
- GenAI — AI that creates new content: text, images, code. ChatGPT, Claude, Midjourney.
- LLM (Large Language Model) — the engine under GenAI for text. GPT-5, Claude Sonnet, Gemini 2.5.
Don't memorise these. You'll absorb them by using them. The vocabulary is just a coat hook — you're hanging concepts on it as we go.
Match the term to its definition.
Quick check. Click a term on the left, then the matching definition on the right. Correct = locks in. Wrong = shake & try again. All 6 = chapter complete.
The T in ChatGPT.
The transformer is the architecture under every modern LLM. It was invented in a 2017 paper called "Attention Is All You Need." The breakthrough: a mechanism called self-attention.
Before transformers, AI systems treated recent words as more important than earlier ones. Self-attention lets the model weigh every word against every other word in the input, no matter the distance.
And actually, the T in ChatGPT literally stands for transformer.— Allie K. Miller · Track 1, Lesson 1
Type something. Watch it become tokens.
Every AI charges by the token, not the word. Roughly ¾ of a word in English. Watch what's actually happening when you type.
When AI confidently makes things up.
Hallucinations are when an AI states something false with the same confidence as something true. Even the top models hallucinate at under 1% — but that 1% can include fake citations, made-up statistics, and "facts" that sound entirely real.
The fix: RAG (Retrieval-Augmented Generation). Instead of relying on the model's memorised knowledge, you give it your own documents to pull from. Now it has to cite what's actually there.
Always verify citations. The link can be real, but the stat or quote inside might not be on that page at all. Verify, verify, verify.
Try this exact prompt in any AI you have open.
This is the cleanest version of every concept you've learned so far. Copy it, paste it into the AI of your choice, see what comes back. Then we'll iterate together in Lesson 2.
Drop this exact prompt into your AI of choice. Test how clear context matters. Bonus: ask a follow-up that pushes the AI to dig deeper.
- You ran the prompt in at least one tool · ideally two for comparison
- You can spot how the analogy lands differently between tools
- You asked one follow-up that pushed the answer further
Three things to take with you.
Pick one. Run it. Save the output to your prompts library.
Pick one by Friday.
-
01
Run the prompt above in two different AIs. Compare. Save the better answer to your
~/promptsfolder. - 02 Tokenise something you wrote last week — an email, a LinkedIn post. Now you know what each one cost when you ran it through AI.
- 03 Spot one hallucination this week. Any AI, any context. Note where it sounded confident and where you'd have been fooled if you hadn't verified.
Save this lesson — the AI/ML/LLM definitions, the S-T-O-L-E-S checklist, and the 4 Cs of safe AI use — into your research notebook.