AI 201 — Under the Hood & Ahead of the Curve

Module 1

How a Machine Actually “Learns”

Inside every model is a neural network: layers of simple math units ("neurons") connected by millions — or billions — of adjustable dials called weights (or parameters). Learning is just tuning those dials. Walk the training loop:

Why "scale" is a big deal: more data + more parameters + more computing = better predictions. That’s why the frontier is a race for chips and energy (the stuff you saw in the Future Quests). Bigger models, trained on more, simply learn more patterns.

Module 2

The AI Landscape

"AI" isn’t one thing. Know the main families you’ll meet:

💬

Language models (LLMs)

Predict and generate text. Power chatbots, writing, coding, tutoring. E.g. ChatGPT, Claude, Gemini.

🎨

Diffusion / generative media

Turn noise into images, video, and audio from a prompt. E.g. Midjourney, Sora, image tools.

🧭

Multimodal models

Handle text + images + audio together — see a photo, read it, and talk about it.

🤖

Agents

Models given tools and goals so they can do tasks, not just chat (next module).

Open vs. closed: some models are closed (used through a company’s app/API — ChatGPT, Claude); others are open-weight (you can download and run them yourself — Llama, Mistral). Open = more control & privacy; closed = usually more powerful and easier.

Module 3

From Chatbot to Agent

A chatbot talks. An agent acts — give a model tools (search, code, a calendar), memory, and a goal, and it can take real steps to finish a task on its own.

Chatbot

Agent

Answers your question

Breaks a goal into steps and does them

Forgets after the chat

Remembers and builds on past steps

Only words

Uses tools: search, code, apps, files

You do the work

It drafts, checks, and iterates for you

RAG (Retrieval-Augmented Generation): the trick that lets AI answer from your documents instead of just its training. It "looks up" relevant info first, then answers — fewer hallucinations, current facts. It’s how most real company AI tools work.

Module 4

Building With AI (You Don’t Have to Be a Coder)

You can create real things on top of these models. The ladder from user → builder:

🧱

No-code

Tools like GPT "custom bots," automation apps, and site builders let you ship an AI product with zero code.

🔌

APIs

A few lines connect your app to a model. This is how developers add AI to anything — the model becomes a building block.

🛠️

Fine-tuning & RAG

Feed a model your own data so it speaks your style or knows your stuff. How companies customize AI.

Portfolio move: the single best thing a graduate can do right now is build and ship one real AI-powered thing — a study bot, a tool for a club, an automation for a job. It beats any certificate.

Module 5

The Hard Problems Nobody Has Solved

The most powerful technology of your lifetime comes with genuinely unsolved problems — and your generation will inherit them. Tap each to understand it.

Why you should care: these aren’t just engineers’ problems. They’re decided by voters, lawyers, writers, founders, and citizens — the roles you’re about to step into. This is the “conscience” layer, and it needs humans.

Module 6

Your Career on the Right Side of the Wave

AI won’t take every job — but it will reshape almost all of them. The strategy is a barbell: work AI can’t easily touch, or work that builds and steers AI. Avoid the easily-automated middle.

🫱

The human end

Hands, heart, trust, judgment — care, trades, leadership, persuasion, craft. Hard to automate, always needed.

🚀

The frontier end

Build and govern the wave — AI, chips, energy, robotics, biotech, and AI policy & safety.

🧭

Judgment

Deciding what’s worth doing — the scarce skill.

🔁

Learning to learn

The tools change yearly; adaptability compounds.

🗣️

Communication

Explaining, persuading, leading humans.

🧩

Be T-shaped

Broad AI literacy + one deep skill you own.

Reference

📖 AI Glossary 2.0

Neural network	Layers of connected math units, loosely inspired by the brain, that learn patterns.
Parameters / weights	The billions of adjustable dials a model tunes during training.
Training	Repeatedly predicting, measuring error, and nudging weights to reduce it.
Agent	A model given tools, memory, and goals so it can take actions, not just chat.
RAG	Retrieval-Augmented Generation — letting AI answer from your own documents.
Fine-tuning	Extra training on your data to specialize a model.
Open-weight	A model you can download and run yourself.
Alignment	Making AI reliably do what humans actually want and intend.
Multimodal	Handling text, images, and audio together.

How a Machine Actually “Learns”

The AI Landscape

Language models (LLMs)

Diffusion / generative media

Multimodal models

Agents

From Chatbot to Agent

Building With AI (You Don’t Have to Be a Coder)

No-code

APIs

Fine-tuning & RAG

The Hard Problems Nobody Has Solved

Your Career on the Right Side of the Wave

The human end

The frontier end

Judgment

Learning to learn

Communication

Be T-shaped

🎓 The Capstone

📖 AI Glossary 2.0