AI Literacy

How does generative AI actually work?

Six tracks. Pick where you are. All of them build toward the same destination: a clear, accurate mental model — no hype, no mystery.

Foundations — What it is and how it works #foundations
Quick Overview · 3 pages · ~15 min
The Generalized Picture
Plain English. No code. An accurate mental model of how AI generates text — in about fifteen minutes.
O1What it is — the pattern machine O2How it learns — wrong to less wrong O3How it guesses — statistics in action
No prerequisites · Start here
Start the Overview →
Deep Dive · 7 sections · Self-paced
The 200 Lines
Walk through a real transformer implementation piece by piece. Every line explained. By the end, nothing is a black box.
§01The problem — what AI is actually doing §02One neuron — the atom of every AI system §03Networks — layers, matrix math, and GPUs §04Training — how a model actually learns §05Attention — the mechanism that changed everything §06The full 200 lines — reading it top to bottom §07So what — applying this understanding
Comfortable with code recommended · Open access
Start the Deep Dive →
Going Further — Scope, limits, and applications #going-further
Beyond Text · 5 pages · Self-paced
When the Language Isn't Words
Images, audio, network traffic, DNA — the transformer doesn't care what the tokens represent. Here's what happens when you change the data.
B1The architecture doesn't care what the data is B2Images as language — patches as tokens B3Audio as language — sound into spectrum B4Network traffic as language — packets as words B5Where else this goes — DNA, proteins, time series
Overview recommended first · Open access
Start Beyond Text →
Right Tool, Right Job · 3 pages · Self-paced
When Not to Use a Text Model
Confident and wrong is worse than obviously wrong. Here's where text models break down — and a practical framework for knowing when to reach for something else.
R1What text models are actually optimized for R2The mismatch problem — images, networks, and data R3How to choose — a practical decision framework
Beyond Text recommended first · Open access
Start Right Tool →
Working With AI — Getting the most out of these systems #working-with-ai
Why Prompting Matters · 4 pages · Self-paced
Context, Context, Context
The model is frozen. You are the only variable. Everything that determines output quality lives in the context window — and you control all of it.
P1The model is frozen — you are the only variable P2Context is the model's entire reality P3What good context looks like P4Advanced patterns — system prompts, RAG, chain of thought
Overview recommended first · Open access
Start Prompting →
Feeding the Model · 3 pages · Self-paced
Injecting What the Model Doesn't Know
The frozen model can't know your data, your documents, or today's news — unless you feed it that knowledge. RAG, search, MCP, and agents are all ways of doing exactly that.
F1The knowledge gap — what the frozen model doesn't know F2RAG and search — retrieving what's missing F3MCP and agents — structured live access to data
Prompting track recommended first · Open access
Start Feeding the Model →