How a model is made.
THE DEFINITION
Adjusting a model's internal values repeatedly until its output matches the training examples well enough.
WHAT IT REQUIRES
Very large amounts of data Substantial computation Time
WHEN IT HAPPENS
Once, before you use the model.
Using the model involves no learning.
WHAT THIS MEANS
The model does not learn from your conversation.
Your input is context, not training, unless the provider specifically uses it that way.
WHY THE TRAINING DATA MATTERS
A model is shaped entirely by what it was trained on.
Its capabilities, blind spots and biases all come from there.
WHAT THE CUTOFF IS
Training data is collected to a point in time. Anything after that is unknown.
RELATED TERMS
Model, parameter, fine-tuning, knowledge cutoff.
