WhatA language model has one job: read the text so far and predict the next small piece of text, a token.
HowIt gives every possible next token a probability, picks one, adds it to the end and runs again on the longer text.
Why it mattersWriting, answering and translating all come from repeating this one loop.