How LLMs Generate Text: Token by Token

Language models don't 'think' and then write. They predict the next token, over and over, using probability distributions shaped by billions of training examples.