Word Embeddings and word2vec

Meaning as geometry

word2vec learns vectors for words such that similar words sit close together — and relationships become directions. The famous example: king - man + woman ≈ queen.

How it learns

By predicting a word from its neighbors (or vice versa) across a huge corpus, the model nudges co-occurring words together in vector space.

Why it still matters

Modern LLMs use contextual embeddings (a word's vector changes with context), but word2vec is the clearest way to build intuition for why embeddings capture meaning — the foundation under all of today's RAG and search.