A concept · How AI Thinks · Computing
Embeddings
Meaning turned into position in a space of numbers.
We type "cheap flights to somewhere warm in February" and the search understands us, though no page on earth contains that sentence. A shop suggests a coat we would actually wear. A colleague's search of the company files finds the right document even though it never uses the word we typed. None of these tools were given a dictionary. How does a machine know that "warm" is close to "Canaries" and far from "Oslo"?
What Embeddings means
An embedding is a list of numbers that places a word, a picture or a whole document at a point in a space, arranged so that things used in similar ways end up close together. "Warm" sits near "sunny" and "Canaries" because those words kept the same company across billions of sentences, and no one told it why. Owning this changes how we search and how we file. We stop hunting for the exact keyword and start describing what we mean, because similarity, not spelling, is what the tool matches. We start noticing when two things the tool treats as neighbours are not neighbours to us, and that gap is usually where its mistakes live. Before: search finds words. After: search finds neighbourhoods, and we learn to ask in their terms.
Not to be confused with
Embeddings are not tokens, and an embedding is not a vector database. A token is a piece of text the model reads. An embedding is the list of numbers that stands for its meaning. A vector database is a place to store and search those lists.
Embeddings examples: where it shows up
- At work: the internal search that finally finds the policy document when we describe the situation instead of guessing its title is matching positions, not words.
- With money: the recommendation that keeps offering us a product we bought once sits in a neighbourhood the model built from other people's baskets.
- In the news: a photo search that mislabels a person has placed two faces close in a space it built, and the story is about who was near whom in the training data.
Where it sits
Concept 1 of 3 in How AI Thinks, a journey in the course Understand AI.
See where Embeddings sits on the sky →
Nearest on the sky
A filled dot marks a star that is already open to walk.
Knowing it isn't owning it.
Connect Embeddings to your mental models and apply it to your world today.
The first one is on us. No signup, no card.