𝗘𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 𝗘𝘅𝗽𝗹𝗮𝗶𝗻𝗲𝗱 𝗦𝗶𝗺𝗽𝗹𝘆
Computers love numbers. They hate meaning.
To a computer, the words "happy" and "joyful" are just different letters. They do not know these words share a feeling.
Embeddings solve this problem. They turn words into lists of numbers. These numbers act like GPS coordinates for meaning.
When you turn words into numbers, similar words land close together in a digital map.
- "Dog" and "puppy" have nearby coordinates.
- "Dog" and "democracy" have far apart coordinates.
A vector is just an ordered list of numbers. "king" → [0.21, -0.44, 0.88] "queen" → [0.19, -0.41, 0.85]
Real models use thousands of these numbers for one word. You do not need to see them all. You only need to know how close two points are.
We use cosine similarity to measure this closeness.
- Points in the same direction = highly related.
- Points at right angles = unrelated.
This turns meaning into geometry. You can even do math with words.
If you take the vector for "king," subtract "man," and add "woman," you land near "queen." The model learns this patterns from reading billions of sentences.
This math powers the AI tools you use every day:
- Semantic search: Finding results by meaning instead of exact keywords.
- Recommendations: Suggesting products or songs based on similar vectors.
- RAG: Helping AI find facts in your documents to answer questions.
- Clustering: Grouping similar items together automatically.
You do not calculate these numbers yourself. You send text to a model and it returns the vector. You then store these vectors in a vector database to search them.
Embeddings turn the mystery of language into the logic of geometry.
Try the Meaning Map to see how words connect: https://dev48v.infy.uk/ai/days/day3-embeddings.html
All concepts: https://dev48v.infy.uk/aifromzero.php
Source: https://dev.to/dev48v/embeddings-explained-simply-how-ai-turns-words-into-a-map-of-meaning-36f4