Representation Geometry
Embeddings place items in a learned vector space.
That lets us inspect distances, directions, and neighborhoods. Items used in similar ways may move toward related regions of the space, because training gives them similar pressures.
The geometry is useful, but it must be read carefully. A coordinate rarely has a simple hand-written meaning. The representation is distributed across many coordinates.
The earlier vector chapters prepared this idea: once items become vectors, similarity and direction become computable.
Embedding geometry is therefore a tool for inspection, not a promise that every nearby pair has a clean human explanation.
Enter 1 if embedding geometry can help inspect learned representations, or 2 if it proves every coordinate has a simple meaning.
Compute it first, then check your number.
In a 2D embedding space, item A is at (0, 0) and item B is at (3, 4). What is their distance?
Compute it first, then check your number.