LLMs work differently, statistically predicting the next token (roughly equivalent to a word) based on all those that came before it, and parameters finetuned during training.

Which is what a parrot does.