There is a strange decision, almost a tasteless one, that I made while putting this living book together: publish the notebook before the essay. Not the clean result that shows up in the talk. The notebook itself, with its crossings-out, its doubts, the prompts that fell off the page and the decisions taken at two in the morning because I could not think about it any longer.
The reason is practical, not aesthetic. Whoever wants to learn to work with artificial intelligence with judgment does not learn from the result. They learn from the attempt. And almost nobody publishes the attempt, because it is uncomfortable, because it exposes you, because it breaks the illusion that brilliant ideas are born brilliant.
Three practices of the open notebook.
One. Document the prompt that failed, not only the one that worked. When a prompt does not produce what you expected, the important thing is not to delete it: it is to understand why it failed. You asked at the wrong emotional temperature. You added context that confused things. You assumed the machine knew something only you knew. Every failure is a lesson the next prompt does not have to learn from scratch.
Two. Capture the detour. The best idea of the session is almost never the one you set out to find. It is one that turned up along the way and annoyed you at first because it was off topic. That is the one worth documenting. It happened to me with a manifesto on artificial intelligence and education: in the detour came a line about institutional silence that ended up being the heart of the project.
Three. Date the contradictions. What you thought on Monday and what you thought on Friday can contradict each other, and that is not a problem: it is process. What is a problem is publishing only Friday and deleting Monday, because then the machinery of thinking becomes invisible.
The process is not what reaches the final page. The process is everything the final page had to forget in order to be readable.
Why publishing this serves other people.
Because when somebody uses the machine for the first time, they assume their clumsiness is theirs alone. That everyone else does it well and only they are struggling. If we only see polished results, that gets confirmed. If we see open notebooks, we discover that clumsiness is the universal condition of the trade, and that the trade is made with exactly that clumsiness, not in spite of it.
That is why this folio is so short, why the logbook carries the failures, and why the Sherezad case tells what broke before it tells what worked. We are going to keep publishing half-cooked things, on purpose. If at some point everything here looks finished, brilliant and without a crack in it, raise your guard: we will have lost the thing.