After tokenization turns your text into numbers, embeddings give those numbers meaning. Every token is mapped to a vector — a long list of numbers that acts like coordinates in a high-dimensional space. Words used in similar contexts end up close together: “cat” sits near “dog,” “Paris” near “London.” The interactive Meaning Map on this page projects real embeddings down to two dimensions so you can see those neighborhoods form.
The most striking property of embeddings is that directions in the space carry meaning. Subtract the “man” vector from “king” and add “woman,” and the nearest word to the result is “queen.” The gender direction, the plural direction, even the capital-city direction all emerge purely from patterns in text — nobody programs them in. The Word Arithmetic demo lets you try these equations yourself with real embedding data.
Embeddings are also the technology behind semantic search and retrieval-augmented generation (RAG): when an app “finds relevant documents,” it is comparing embedding vectors. Understanding this page is understanding half of how modern AI products work.
Embeddings power half the AI tools you use. Get my AI roadmap
howaiworks.io is free and open source (GitHub), built by Matt Feroz.