Embedding
A numeric vector representation of text (or images, audio) where semantically similar items are close together.
An embedding turns a piece of content into a list of numbers (a vector) such that similar content has similar vectors. 'King' and 'queen' end up close in embedding space; 'king' and 'banana' do not.
Embeddings power semantic search, recommendation, deduplication, classification, and most importantly retrieval-augmented generation. Modern embedding models like OpenAI's text-embedding-3, Cohere's embed v3, and Voyage AI's voyage-3 produce vectors of 1024-3072 dimensions.
The quality of your embeddings caps the quality of any RAG system you build on top. Specialized embeddings (code, multilingual, multimodal) often outperform general-purpose ones for narrow tasks.