Understand
the next word.
How does AI turn your question into an answer? Follow it, one small step at a time.
Start with a short lesson, then try an example yourself. No coding or maths needed to begin.
First lesson: What is a token? · 6 min
Start with a question.
A familiar question is a good way into the machinery. Pick one to find a short explanation.
Why does AI make things up?
Plausible language and reliable evidence are different things.
Understand hallucinations →Why do I get different answers?
The same candidate scores can lead to different sampled tokens.
Explore how choices are made →Does it remember our conversation?
Separate the current context from the model’s learned weights.
See what context means →How does a model learn?
Follow predictions, errors, and repeated weight updates.
Follow the training loop →Can it look things up?
See how a surrounding system supplies external sources.
Understand retrieval →Why does it split up words?
Meet the pieces that turn text into a sequence of numbers.
Get to know tokens →What would you like to understand?
New to language models? Start with The big picture. Each path puts the lessons in order.
The big picture
Build a useful mental model without starting with the mathematics. Follow the path from text to a response, then learn where that picture stops.
Follow a token
Travel through the numerical pipeline, from text pieces to a probability distribution. Use the worked pipeline alongside the lessons.
How learning happens
Connect data, prediction error, gradients, and updates. Make one small weight learn before considering a large training run.
Learning is one process.
Responding is another.
Training changes the model’s weights. Inference uses those weights with the context you provide. A chat product may add tools, retrieval, and more.
Trace both flows →Change a number. See what changes.
Eight transparent experiments, running right in your browser.