One question has always bothered me. How do these language models know who or what they are? For example, When you ask it questions and it gives responses lik

One question has always bothered me. How do these language models know who or what they are? For example, When you ask it questions and it gives responses like "I'm an AI", or it may know it's name. So how does the language model know what it is from the training data? Is it somehow hard-coded in, or are there specific bits of training data used to make that connection for it? Could you make a language model that did not know what it was?