Parametric knowledge
What a model “knows” baked into its weights, not looked up from anywhere.
Think of it like
Like facts you’ve memorized versus ones you’d Google — parametric knowledge is the stuff already in your head.
Example
A model naming the capital of France with no documents provided is drawing on parametric knowledge.
How it actually works
During training, facts and patterns get compressed into the parameters themselves. It’s fast and always available, but it’s frozen at the knowledge cutoff, can be subtly wrong, and can’t cite a source. That’s the trade-off retrieval addresses by pairing it with non-parametric memory.
For product teams
Convenient but unverifiable — fine for general knowledge, risky for facts that must be current or exact.
For engineers
Information encoded in model weights, accessed without retrieval; static, uncited, cutoff-bound.
Related
- Non-parametric memory — Its looked-up counterpart.
- Knowledge cutoff — The date it freezes.
- Weights — Where the facts live.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome