Training is the learning stage in which an AI system uses many examples to recognize patterns and later produce outputs for new inputs. The costly work can happen once, then support many predictions instead of being repeated every time. section 11.6 section 1.5
A practical way to understand training is to separate it from later use:
- Training uses a large collection of examples, such as millions of open-source repositories. section 1.5
- After training, the system can generate code or make predictions without training from scratch for every request. section 11.6
- Training data can contain older practices, so generated output still needs checking. section 2.6.1
Checking trained knowledge against Python 3
Copilot may suggest the older form print "Hello Copilot" because its training included old Python code. In a Python 3 environment, replace it with print("Hello Copilot") and test the result. The same checking habit applies to generated code for tasks such as calculating a square's area. section 2.6.1
