[sp332]

It's just a guess, based on the LLM never getting writers block and being able to hold a good amount of context in mind for every output token.