Most gardens are wandered, not read front to back. This is a map of the main paths.
How to read a garden
Follow the links, not the order. Every
[[wikilink]]is a doorway; hover one to peek inside before you step through.
Start here: the five-part path
- The atoms of a model
- Embeddings, explained
- How a transformer thinks
- Temperature and the next word
- Watch a model learn
Zoom out with the whole model on one screen, or get hands-on with teach a neuron.
Concept clusters
- The input side: tokens, embeddings, vector databases, context window.
- The model: neural networks, transformers, attention, softmax.
- Using models: prompts, temperature, RAG, fine-tuning, hallucination, LLMs.
Tools and protocols
The browser tools (token & cost calculator, tokenizer comparison) and the protocol explainer, What is MCP?.