What happens each time an LLM writes a word
How does an LLM decide what to write next?
An LLM writes one small piece of text at a time. It reads everything so far, gives every possible next piece a chance, picks one, adds it, and goes again. Here is that loop on a real model, with its real numbers.
