Parameters
The learned numerical values inside a model that encode patterns acquired during training.
Illustrative conversation
- Can we edit one parameter to correct a single outdated fact?
- Not reliably. Knowledge is distributed across many learned weights rather than stored as one editable record.
Definition
Parameters are the numerical weights and biases a model learns during training. Together they shape how input representations are transformed into predictions, encoding statistical patterns rather than a neatly addressable database of sentences or facts. Parameters normally remain fixed during inference; changing them requires additional training or a parameter-update technique such as fine-tuning.
Reviewed