AI Terminology Course
AI Terminology
/
Advanced

Embeddings

Definition

The mathematical translation of concepts (words, sentences, or images) into high-dimensional arrays of numbers (vectors) in such a way that similar concepts are located close to each other in mathematical space.

Explain Like I'm New

Giving words GPS coordinates. The AI plots the word 'King' on a map. It plots the word 'Man' nearby. It plots the word 'Apple' a million miles away. If you tell the AI 'Find me words near King', it looks at the map and finds 'Queen' and 'Prince'.

Real World Example

Spotify uses embeddings. It turns every song into a mathematical vector. When you listen to a Rock song, Spotify's algorithm searches the 'Vector Map' for other songs with very similar mathematical coordinates and puts them in your Discover Weekly playlist.

Common Use Cases

  • •Semantic Search
  • •RAG systems
  • •Recommendation engines

Interview Questions

basic

  • Do Embeddings represent words as text strings, or as arrays of numbers?

intermediate

  • What is 'High-Dimensional Space' in the context of Embeddings?

Flash Cards

Question

Strings or numbers?

Click to reveal answer
Answer

Arrays of floating-point numbers (Vectors).

Question

What is High-Dimensional Space?

Click to reveal answer
Answer

A standard map has 2 dimensions (X and Y). A 3D graph has 3 dimensions (X, Y, Z). OpenAI's embeddings plot words in 1,536 dimensions! Each dimension represents a highly abstract concept (like 'royalty', 'gender', 'fluffiness') allowing the AI to perfectly capture the nuance of a word.