Parametric knowledge is everything a model knows by heart: the facts, patterns, and skills baked into its parameters during training. When you ask a question and the model answers without looking anything up, it is drawing on parametric knowledge.
Broad, fast, and unreliable
This is the model's built-in memory, and it is genuinely vast: most languages, common libraries, and general programming wisdom, all sitting in the weights and instantly available with nothing loaded into context. That is the strength.
The weaknesses come from the same source:
- Frozen. It stops at the knowledge cutoff, so anything newer is missing or wrong.
- Unsourced. The model cannot cite where a fact came from, and cannot tell a real memory from a plausible guess, which is where hallucination creeps in.
- Blurry on specifics. It is strong on the shape of things, shakier on exact version numbers, obscure signatures, and your particular codebase.
When to lean on it, when not
Parametric knowledge is fine for the general and stable: how a for-loop works, what HTTP is, the idiom for reading a file. It is the wrong thing to trust for the specific and current: a library's latest API, your internal conventions, this week's schema. For those, override memory with contextual knowledge by putting the real source in front of the model.
Related terms
Training
Training is the process that produces a model: showing it enormous amounts of text and adjusting its parameters until it gets good at predicting what comes next. It happens once, before you ever use the model.
Read definition →Knowledge cutoff
The knowledge cutoff is the date after which a model learned nothing from training. It is a common source of outdated APIs, so give the model current docs to compensate.
Read definition →Contextual knowledge
Contextual knowledge is what a model knows because it is in the context right now: the files, docs, and output you gave it. It is current and grounded, and it is the main lever against hallucination.
Read definition →