Words people use about AI, in plain English
One paragraph each. If a word on this site is not here, tell us and we will either explain it or stop using it.
What is a model?
A machine that guesses what comes next, built by showing it an enormous pile of examples. When someone says AI, this is usually the thing they mean.
What is training?
The months of expensive computation that built the model, before anyone used it. It is finished by the time the model reaches you. When a vendor says the model learns from your data, ask them which of these two things they mean, because usually they mean neither.
What is a hallucination?
A confident answer that is simply invented. It is not a fault in the machine. A machine built to guess will guess, and it has nothing in it that knows the difference between a guess and a fact.
What is AI governance?
Being able to answer four questions about any answer the system gave: which data it read, who asked, what it said, and when. Most of what people call governance is a document. This is the part that is a working system.
What does it mean when a model looks things up?
Instead of relying on what the model absorbed while it was built, the system finds the relevant documents first and hands them over with the question. The answer can then cite which document it came from, which is the whole point. Vendors call this RAG.