Latent space
The hidden map where the model arranges everything it knows, with similar things placed near each other.
Think of it like
A vast library where books shelve themselves by meaning, not by title — cookbooks drift together without anyone labeling the aisle.
Example
In a good latent space, “king” minus “man” plus “woman” lands you near “queen”.
How it actually works
“Latent” means hidden: it’s the internal coordinate system a model learns, where position encodes meaning. You never see it directly, but retrieval, clustering, and analogies all work because related concepts cluster there. The geometry is learned, not designed.
For product teams
It’s why semantic search “just knows” two differently-worded things mean the same — they sit close in this space.
For engineers
The learned embedding manifold; a continuous vector space where proximity reflects semantic or feature similarity.
Related
- Vector — Made of vectors.
- Dimensionality — Its width is dimensionality.
- Embeddings — Points in it are embeddings.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome