What it means
Embeddings are lists of numbers that represent the meaning of a piece of text, an image or other content, created by an AI model. Content with similar meaning gets similar numbers, which allows computers to find related content even when different words are used.
Real-world example
In a city's knowledge base, the guidance page "Registering a new address" and a citizen's question "I've just moved, what do I need to do?" share almost no words. Their embeddings, however, are close together, so the system can recognise that the page answers the question.
Why it matters for AI adoption
Embeddings make semantic search and most RAG systems possible. The choice of embedding model affects how well the system understands specialist vocabulary and different languages, such as German administrative terms. Changing the embedding model usually means re-processing all documents.
Often confused with
Embeddings are the numerical representations of content. A vector database is the system that stores large numbers of embeddings and finds the most similar ones quickly. You need an embedding model to create embeddings; you need somewhere to store and search them.
Questions to ask
- Which embedding model do we use, and how well does it handle German and our specialist terms?
- What would it take to switch to a different embedding model later?