StudyAIStudyAI
Pro
Lesson 28 min

Tokens & Embeddings

How models turn words into numbers they can compute on.

What you will learn
  • What a token is
  • What embeddings represent
  • Why this matters for cost and search

Explanation

Models do not see words — they see numbers. Text is first split into tokens (roughly word-pieces; 'studying' might be 'study' + 'ing').

Each token is mapped to an embedding: a list of numbers (a vector) that captures meaning, so similar words sit close together in that number-space.

Tokens matter practically: APIs charge per token, and context limits are measured in tokens. Embeddings power semantic search and RAG.

Real-world use

Search that understands meaning ('cheap laptop' matching 'budget notebook') uses embeddings, not keyword matching.

Common mistakes
  • Confusing tokens with words — a token is usually smaller than a whole word.
Practice

Estimate how many tokens a 100-word paragraph is (hint: ~1.3 tokens per word).

Knowledge check
0/2 answered

1. An embedding is...

2. Why do tokens matter practically?

Answer all questions to check.