Parametric Knowledge
Pengetahuan Parametrik
What the model knows from training, stored in its parameters. Frozen at training time.
Parametric knowledge is what the model knows from training, distributed across its parameters in a form nobody can read as rules. It is frozen at training time and never updated afterwards, which gives it a hard boundary: everything up to the knowledge cutoff, nothing after it.
This is the knowledge that is there but not available on demand. You cannot ask a model to show you where it learned something, and you cannot remove a fact from it. It is also why the same question can be answered confidently and wrongly, because the model is retrieving a statistical association, not looking something up.
Design implication: parametric knowledge is good for general fluency and bad for anything specific and current. When a fact matters, do not trust the parameters to supply it. Put it in the context, where it is checkable, and treat the model's own memory as a hint rather than a source.
Related terms
Contextual Knowledge
Facts the agent can read directly from the context right now. Counterpart to parametric knowledge.
Knowledge Cutoff
The date after which the model has no trained knowledge at all.
Parameters
The learned numbers inside a model. Their count is the model size.
Hallucination
Output that is fluent and wrong. Two different problems hide under one word.
Grounding
Tying the answer to something checkable, so it can be verified instead of trusted.
Want this working inside your team?
It starts with one conversation about your workflow.