Learn
Two questions, answered with pictures.
Most people think AI is one thing. It is not. This page breaks AI into the parts that matter. The first half shows you how AI thinks. The second half shows you how people build real AI systems. Read it in order, or skip to what you need.
How AI Works
From spam filters to ChatGPT, step by step.
- 1RULES-BASED
- 2CLASSICAL ML
- 3NEURAL NETWORKS
- 4LLMs
- 5TRAINING
- 6WHY LLMs DIFFER
Rules-Based Systems
(the baseline)
A human writes the rules. The computer follows them.
Classical ML
(the pattern finders)
The computer learns rules from examples. No human writes them.
Neural Networks
(the layered thinkers)
Stacks of small pattern finders, working together.
LLMs (normal AI)
(the context machines)
The thing ChatGPT is. It guesses the next word, very well.
Training
(how they learn)
A guessing loop that runs a trillion times.
Why LLMs Feel Different
(the payoff)
One trick does translation, writing, and code.
How AI Systems Get Built
The parts people add to LLMs to make them useful.
LLMs Alone
(the raw ingredient)
Just the model. No memory of your files. No tools.
RAG
(giving it a textbook)
The model reads your documents before it answers.
Tools
(giving it hands)
The model can call code you wrote. Check the weather. Send an email.
MCP
(standardising the hands)
One plug that fits any tool. No custom code per app.
Agents
(letting it work unsupervised)
The model runs in a loop. Works for minutes, not seconds.
Everything you have seen AI do in the last two years is some mix of the things on this page. There is no twelfth thing. Once you see the parts, the magic turns into engineering.