Tutorials · Chapter C (3/4) · ~7 min
Embeddings — meaning as numbers
Play → read → next
Park word-planets near their meaning neighbors.
Simulation game
Embedding galaxy
Select a word-planet, then tap the night sky where its meaning belongs.
Recap
What you just did
You placed ideas on a map so similar meanings sit nearby. That map is the intuition for an embedding: a meaning fingerprint made of numbers.
Teach
How it works
An embedding model turns text into a list of numbers (a vector). In that space:
- “puppy” lands near “dog”
- “joyful” lands near “happy”
- “toaster” lands far from “king”
Apps use distance: nearest neighbors = related meaning. That’s the engine behind many recommenders, semantic search, and RAG.
Use it
When you'd use this
- Search notes by idea (“refund policy”) even if the doc says “money-back”
- Group feedback themes without hand-labeling every phrase
- Find “similar lessons” or related support tickets
Watch out
Watch out
Embeddings are fuzzy. Nearby ≠ identical. Always read the retrieved chunk before you trust it.
Try next
Try this next
List three phrases that mean “I’m angry.” They should embed nearer to each other than to “I’m hungry.”