The learning hub

You can understand this.

Neural networks do not have to be a black box. Explore the ideas one piece at a time, at your own pace.

START HERE / INTERACTIVE

The next word.
A bigger idea.

How do small pieces of text become a whole answer? Play with probabilities and look inside the process.

Try it yourself
INPUT TEXT
The modelselectsthe next
↓ WHAT COMES NEXT?
tokenwordanswer

Twelve ideas, one clearer picture

FOUNDATIONS · APPLICATIONS · CRITICAL THINKING
01 / TEXT & LANGUAGE

How a neural network picks the next word

Tokens, probabilities and temperature. Explore the idea with a hands-on example.

Interactive
02 / INSIDE THE MODEL

Attention: connecting the words

How a model uses context, and why word order matters.

Explainer
03 / IMAGES

How noise becomes an image

What happens between random dots and a finished picture.

Explainer
04 / FOUNDATIONS

Training and inference: learning, then answering

What changes inside a model during training — and what happens when you send a prompt.

Explainer
05 / FOUNDATIONS

Embeddings: meaning as a map of numbers

Why a search can find related ideas even when the words do not match.

Explainer
06 / BUILDING WITH AI

RAG: let the model look things up

Retrieval-augmented generation connects an answer to documents you can inspect.

Explainer
07 / BUILDING WITH AI

AI agents: when a model can take a next step

Tools, feedback loops and permissions — what turns a reply into an action.

Explainer
08 / INSIDE THE MODEL

Reasoning models: spending more work on an answer

Why extra inference-time computation can help — and why a longer answer is not proof.

Explainer
09 / BUILDING WITH AI

Fine-tuning, prompting or RAG?

Three ways to adapt a system, each changing a different part of the process.

Explainer
10 / RUNNING MODELS

Quantization: fitting a model into less memory

What 4-bit weights mean, and why downloading the file is only part of running a model.

Explainer
11 / READING THE NEWS

Benchmarks: what does a higher score really tell you?

Read model comparisons through tasks, test conditions and practical trade-offs.

Explainer
12 / INSIDE THE MODEL

Context windows: room to read is not perfect memory

What fits into a request, what uses the budget, and why important details can still be missed.

Explainer