Skip to main content

Learning the Rows

This video presents the same text shown beside it, spoken and on screen. It adds nothing the text does not say.

State

Real embeddings are learned — descent adjusts a few hundred numbers per word so that prediction error falls, and the resulting geometry organizes more than synonymy.

Show

The word2vec family is the classic recipe; directions in the learned space track relations the training never named.

Watch for

Nobody typed the geometry in — the rows sit wherever prediction pressure pushed them.

Builds on

Unlocks

  • Nothing yet depends on this.