Let's uncover how transformer models sitting behind LLMs analyze input information like user prompts and how they generate coherent, meaningful, and relevant output text "word by word".