AI glossary for decision makers
Five terms are enough to decide. The rest can wait.
Join our Master ClassYou do not need to know it all
To judge an AI project you do not need a full glossary. Five terms are enough to ask the right questions and to notice when a supplier is dodging them.
The rest becomes useful as you go further. That is what the full glossary, with 84 terms, is for on our technical site.
RAG
The technique where the AI first searches your own documents and only then answers. Without RAG a language model invents; with RAG it cites. That is the difference between a conversation and a usable answer.
LLM
A language model of large size, such as Mistral, Claude or GPT. It is the engine that understands and produces language. Which model you pick changes the cost, the quality and where your data sits.
Sovereign AI
An AI where you control the tooling, the data and the place of processing. The test is simple: if it stops without internet, it is not yours.
On premises (local)
Software running on your own hardware in your own building, as opposed to the cloud. Dearer to start, cheaper over time, and your documents never leave.
Token
The unit cloud services bill by: roughly three quarters of a word. Processing a thousand pages costs thousands of tokens, on every run. Locally it costs nothing.
The full glossary
Eighty-four terms across seven areas: infrastructure, databases, artificial intelligence, development, version control, orchestration and method. Each with a concrete comparison.
See the full glossaryWould you rather have someone explain it?
Thirty minutes with an engineer is worth more than thirty glossaries. We will tell you what matters in your case, and what you can skip.
A 30 minute call