PART 1 · MODULE 3How AI understands meaning
Embeddings
Turning words into numbers that capture meaning
Loading module...
Turning words into numbers that capture meaning
AI models convert words into lists of numbers called embeddings. These numbers capture the meaningof a word — so words with similar meanings end up as similar numbers. Think of it as giving every word a GPS coordinate in a "meaning space."
Words as points in space — similar meanings cluster together
Each word becomes a vector — a list of hundreds of numbers (e.g., 384 dimensions).
"budget" → [0.12, -0.34, 0.87, ...]Words with similar meanings are placed close together in this number space.
"budget" and "expenses" are close neighborsYou can compute similarity between any two words using cosine similarity.
similarity(budget, expenses) = 0.85Add words and watch them appear in 2D embedding space. Click any two words to calculate their cosine similarity. Uses a real embedding model (all-MiniLM-L6-v2) running in your browser.
Search by meaning, not just keywords. "How do I request time off?" finds the PTO policy even if it never says "time off."
Embeddings power retrieval — they're how RAG systems find the most relevant document chunks for your question.
Group similar emails, tickets, or documents automatically by comparing their embeddings.
"People who liked X also liked Y" — because X and Y have similar embeddings.
Imagine your company org chart, but for words. Embeddings place "quarterly report," "financial summary," and "Q3 results" in the same neighborhood — even though they use completely different words. That's how AI understands that a search for "budget overview" should also find documents titled "fiscal plan."