A language model is trained to predict the next word in a sequence of words. Thus the "programming" is just a prompt that primes the model in a certain way. There is no difference between user input and prompt - the training data doesn't allow it to make that distinction.