Semicolony ELI5 · comic

Embeddings.

Turning words and images into points on a map, so things that mean the same thing sit close together.

  1. These two? Total strangers.
    car automobile nothing in common
    1

    To a computer "car" and "automobile" are just letters with nothing in common.

  2. car → a row of numbers.
    car model 0.8 0.1 0.4
    2

    An embedding model reads each word and turns it into a list of numbers.

  3. Numbers are a spot on the map.
    car a spot on the map
    3

    Those numbers are really coordinates — a single point dropped on a map of meaning.

  4. Synonyms? Right next door.
    car · auto banana far apart
    4

    Trained on mountains of text, the model puts similar meanings near each other.

  5. Closeness equals meaning.
    close = similar far = unrelated
    5

    Now meaning is just distance: close points mean similar, far apart means unrelated.

  6. Different words, same spot.
    reset my password account recovery same spot on the map
    6

    That powers search by meaning: "reset my password" finds "account recovery."

Words become points on a map, where things that mean the same thing sit close together.
Semicolony semicolony.dev/eli5/embeddings/comic
← All ELI5 explainers