How LLMs Generate Text — one token at a time

One token at a time — a very well-read autocomplete. A large language model doesn't plan a whole answer up front. It predicts the next token from everything so far, appends it, and repeats — with atte...