Knowledge cutoff
A model's knowledge cutoff is the date past which it has no built-in knowledge. A model only "knows" what was present in its training data, and training stops at a fixed point in time. Anything that happened, changed or was published after that point is invisible to the model unless you provide it.
This is a frequent source of confident mistakes. Asked about a recent event, a new product or a policy that changed last month, a model will often answer anyway, drawing on outdated information or filling the gap with a plausible guess rather than admitting it does not know.
A model rarely says "that is after my cutoff." It tends to answer with the same confidence either way, so an out-of-date answer can look exactly like a current one. Treat questions about recent or fast-changing facts with extra caution.
Why it matters in Rational AI
The knowledge cutoff is one of the main reasons the platform does not rely on a model's built-in knowledge alone. Instead, it supplies current, organisation-specific information at the moment of the request:
- Your Knowledge holds your own, up-to-date sources, independent of any model's training date.
- RAG retrieves the relevant pieces and places them in the context window, so the model answers from your material rather than from stale memory.
This is the distinction between what a model carries in its weights and what it reads at request time, covered in parametric and contextual knowledge.