Definition
A specialized neural network used in RAG to transform text chunks into high-dimensional numerical vectors (embeddings) that represent semantic meaning. It enables mathematical similarity comparisons during retrieval, where a higher dimensionality usually yields better precision at the cost of increased latency and storage requirements.
Not the LLM that generates text, but the encoder that converts text into coordinates.
"A GPS coordinate system for ideas where 'apple' and 'fruit' are located on the same street, even if they share no letters."
- Vector Database(Component)
- Cosine Similarity(Mathematical Metric)
- Chunking(Prerequisite)
- Semantic Search(Use Case)
Conceptual Overview
A specialized neural network used in RAG to transform text chunks into high-dimensional numerical vectors (embeddings) that represent semantic meaning. It enables mathematical similarity comparisons during retrieval, where a higher dimensionality usually yields better precision at the cost of increased latency and storage requirements.
Disambiguation
Not the LLM that generates text, but the encoder that converts text into coordinates.
Visual Analog
A GPS coordinate system for ideas where 'apple' and 'fruit' are located on the same street, even if they share no letters.