04 / How It Goes Wrong
Contextual knowledge
What you put in front of the model right now. Narrower than training knowledge, but current and checkable, and therefore almost always the more trustworthy of the two.
How you’ll hear it
“Can we trust this answer?”
“More than usual. It is working from the document we gave it.”
Avoid
Mixing supplied documents and general knowledge without saying which is which.
The longer version
Contextual knowledge is what the model reads from the material you provide right now. It is current and checkable, which makes it the more trustworthy source. Good systems make the model rely on it and show where each claim came from.