7 minute interactive lab

Embeddings

How does meaning become geometry?

YOU’LL LEAVE KNOWINGEmbeddings encode useful relationships as directions and distance.

THE EXPERIMENT

Navigate a map made of meaning.

Move the controls. Watch the internal state respond. The explanation follows the behavior—not the other way around.

Conceptual simulationThe behavior is deliberately simplified and labeled; it teaches the relationship, not a trained production model.

VECTOR SPACEClick a word to inspect its neighborhood

A model does not store a dictionary definition. It learns coordinates that make useful relationships nearby.

Nearest to king

01queendistance 16

02mandistance 23

03womandistance 28

king − man + woman ≈queen

royalty →technology →

THE MENTAL MODEL

Embeddings encode useful relationships as directions and distance.

An embedding is a learned list of coordinates. Training arranges those coordinates so that words used in similar ways become neighbors and useful relationships become directions.

IN DEVELOPER TERMSsimilarity(a, b) = cosine(vector_a, vector_b)

GO DEEPER

Check the model against the source.

Connectionism optimizes for intuition, then points you to the rigorous treatment.

NEXT CONNECTION

Which tokens should influence this one?

The attention mechanism

Continue to lab 010 →