concept
A new model starts out knowing nothing useful. During training it is shown a very large number of examples and makes predictions; each time, its error is measured and its parameters are nudged slightly to reduce that error.
Repeated an enormous number of times, those small nudges add up to a model that has captured patterns in the data. Training large models takes a great deal of data and computing power.
When training ends, what the model has learned is fixed, unless it is trained further or given new information in its prompt.
Example
A model trained on many labelled photos of cats and dogs learns to tell them apart in photos it has never seen.