Training Data
1 min read · Reviewed 17 Oct 2026
In short
The material a model learned from — and the limit of what it knows.
In plain English
Everything a model can do comes from what it saw during training: text, code, images, whatever the developers fed it. It has no other source of knowledge unless a tool gives it one at the moment you ask.
A real example
Ask a model about a well-documented topic and it performs well, because it saw a great deal about it. Ask about your industry's niche practice and it may be confidently vague — there simply was not much to learn from.
Why it matters
It sets the boundary of what to trust. Well-documented, widely written-about, historically stable: probably fine. Niche, recent, local, or contested: verify it.
