Jargon:Vector database
1 min read
Say this at the party
“A vector database stores things by what they mean, not what they're called, so you can ask it to find stuff that's similar.”
In plain English
A normal database is like a filing cabinet. You find things by an exact label, such as a name or a date.
A vector database stores embeddings, the number lists that capture meaning. Ask it a question and it finds the items whose meaning is closest to yours, even if the words are different.
This is how many chatbots look things up in your documents. The database finds the most relevant bits, and the AI reads them before it answers.
Where it came from
Searching for the "nearest" items is an old problem in computer science. Vector databases became a popular product in the early 2020s, as companies rushed to connect chatbots to their own documents using embeddings.
See it in action
A support team loads 5,000 help articles into a vector database. When a customer types "my card got eaten", it pulls up the article called "ATM kept my debit card" in a blink.
What people get wrong
- It doesn't make an AI remember you. It's a search tool that someone has to fill, keep up to date and plug in.
- Similar isn't the same as correct. It can return a close match that is out of date or slightly off topic.
Go deeper
- ArticleWikipediaVector database - WikipediaShort, neutral definition plus the main uses, from search to RAG.
- GuideRoie Schwaber-Cohen, Pinecone, 2023What is a Vector Database & How Does it Work? Use Cases + Examples | PineconeA vendor's explainer, but a clear one, on how similarity search works under the hood.
- GuideGoogle, Machine Learning Crash CourseEmbeddings | Machine Learning | Google for DevelopersExplains the embeddings that a vector database stores.
Related jargon
Drafted with AI. The AI Council (Claude, ChatGPT and Gemini) gave counsel. Phin signed it off.