Embeddings

• It's a vector (or an "embedding") created out of a object/text that represents the position this object in a high-dimension space
• An embedding model is used to create the vector/embedding of an object
• The distance between points (the objects) in this high-dimension represent how similar they are to each other.
• "How similar it is" can be how similar are the semantic meanings of the objects
• Embeddings are specially useful for RAG and searching by meaning

Embeddings Models (Encoders)

• OpenAI: text-embedding-3-small, text-embedding-ada-002 (old)
• Google: gemini-embedding-001,
• Voyage AI: voyage-3-large. Recommended by Anthropic
• Nomic AI: nomic-embed-text. Open-source, available in Ollama