Real-time thought development
The text shows traces of how the author reaches a conclusion or corrects an opinion right while writing.
"While working on the old landing page, we encountered layout issues. However, after conducting additional analysis and replacing outdated elements with modern solutions, we successfully restored proper display. Thus, refactoring proved inevitable."
"Yesterday evening I sat down to poke at an old landing page for a client (I think it was a construction firm, or window repairs, doesn't matter). I just needed to fix some margins. I ended up stuck for three hours because the whole layout broke... Anyway, I gave up and rewrote the block using CSS Grid."
The foundation of human text
A neural network (LLM) generates an answer from left to right, token by token, but it statistically "knows" the structure of its conclusion before writing the first word. Any of its U-turns like "however" are part of a pre-calculated plan.
For a human, writing is a rough draft that goes straight to print. The author tries to recall a detail, realizes in the process that it's irrelevant to the story ("doesn't matter"), and captures this process on the fly. They leave their entire thought path visible, without any theatrical "wait, I changed my mind".