Embedding
An embedding is a way of representing something, such as a word, sentence, or image, as a list of numbers that captures its meaning. Things with similar meanings end up with similar numbers, so a computer can measure how related two items are by comparing their embeddings. This quiet piece of machinery is what lets modern search understand intent rather than just matching keywords, and it is central to how systems like RAG find relevant information.
Computers cannot compare meanings directly, but they are excellent at comparing numbers. An embedding bridges this gap by placing each item at a point in a vast, invisible space of numbers, arranged so that similar things sit close together. In such a space, “king” and “queen” would be near each other, “dog” and “puppy” would be neighbours, and “banana” would be far from all of them. The computer never understands meaning the way we do, but it can now measure closeness, which turns out to be remarkably powerful.
Picture arranging books in a library not by title but by subject, so that everything about British history clusters in one area and everything about cookery in another. Ask for something similar to a book on Tudor kings, and you look nearby rather than hunting through every shelf. Embeddings do this for words, documents, images, and more, giving each a position based on meaning so that “find me something like this” becomes a matter of looking at what is close by.
Embeddings underpin a huge amount of everyday AI. They power search that copes with synonyms and phrasing, recommendation systems that suggest similar products, and, crucially, retrieval-augmented generation, where a system fetches the passages most relevant to your question before answering. Whenever an AI product seems to grasp what you mean rather than just the exact words you typed, embeddings are usually doing the work behind the scenes.