Are the LLMs in any way trained semantically or by hooks that you can plug in, say, Python docs? And if a new version of Python then gets released then the trai

Are the LLMs in any way trained semantically or by hooks that you can plug in, say, Python docs? And if a new version of Python then gets released then the training data changes, etc