Explorable AI.
Type a sentence and watch it travel through a language model, tokens, embeddings, attention, and its guess at the next word, all recomputing as you type. Hover any token to trace it through every stage at once. Nothing below is a diagram; it is the thing running.
That is the whole idea of this site: learn how modern AI works by playing with it, not hype, not equations for their own sake, but ideas you can touch. Every explainer has a live widget like the one above. Stuck on anything? The ✦ Ask AI button in the corner explains it right on the page, and there are free tools you’ll want to keep around.
Start with the five that build on each other:
- The atoms of a model, watch your words dissolve into tokens.
- Embeddings, explained, where meaning becomes distance on a map.
- How a transformer thinks, click a word and see attention light up.
- Temperature and the next word, load the model’s dice and watch it get weird.
- Watch a model learn, roll the ball downhill; that’s training.
Want the big picture first? See the whole model on one screen (every stage lit up at once) or teach a neuron by drawing your own data.
Building with the API? The token & cost calculator counts any prompt exactly (using each model’s real tokenizer) and estimates what a call costs across models. It runs entirely in your browser.
Or ask the garden, type a question and an embedding model running in your browser finds the notes closest in meaning (a live RAG demo):
Prefer to wander? The concept notes are a linked encyclopedia of AI terms. Start anywhere and follow the threads, or open the graph view.
Or explore the whole garden by meaning. The graph view links notes that reference each other; this map places every note by what it’s about, a mini dimensionality-reduction of the garden itself. Hover a dot to see its closest kin; click to open it.
And here’s the garden being tended, my commits over the last year, pulled live from GitHub:
- Oliveaniss