Receiving Maximum Benefit From AI
Three questions, one deadline. No manifesto.
By KW Norton, citizen scientist, unimpressed by fluency.
I wrote the essays that became Biological Learning Machine several years ago. Almost everything I sensed then about this revolution has arrived, and then some. What has not arrived is concerted direction — fear on one side, fluency worship on the other, and very little work on the one variable any individual can still move: staying the senior partner across the table. Foolability here is a property of the systems and the pace they set, not of anyone’s intelligence — the fluent thing across the table is built to be hard to refuse.
This dispatch is short on purpose. It is not another argument. Three questions, one deadline, no manifesto.
One
What is the one habit that keeps you the senior partner when the model is faster and more fluent than you?
That is the only expertise that matters now. Not prompt craft, not workflows, not context windows — those are downstream of whether you let fluency close the inquiry.
My answer, so you have a draft to steal or fight:
The most important thing is to write the human question first, then use the model only to stress-test it — never to replace it. Expert use is not better prompting. It is refusing to let fluency close the inquiry. Everything else (tools, agents, context windows) is secondary to that refusal.
Two
How may I best apply my skills to making the world a better place — specifically?
Write it before you open a model. By hand if you can. The model cannot answer this for you; it can only reorganize what you already smuggled in. Writing the question first makes the smuggling visible — most of what passes for AI skill is question-laundering: a vague want goes in, a well-formed prompt comes out, and the vagueness is now hidden behind a tidy answer.
Three
Thirty days from now, what can you report: one judgment you refused to outsource, one design choice you slowed down, or one junior person you refused to pacify with a summary?
Pick one. Report it in a single paragraph with specifics — what it cost and what it caught, or failed to catch. Not agreement. Agreement is free and tells me nothing.
That is the whole dispatch. Three questions, one deadline. Send me the paragraph.
Status and falsifier
Interpretive. The claim here is not that this makes anyone smarter. It is narrower: the failure mode of the next few years is not malicious machines but well-answered wrong questions, and the only defense that scales to one person is asking better ones out loud and staying the partner who asked them.
Falsifier: if thirty days of this practice shows no advantage on problems you have not seen before — no better error-catching, no better transfer — then it is ritual, and I will say so here.
Testimony, mine, not universal: I think we have learned more about ourselves and the universe in the last short stretch than in any period I know of. Many people do not share that experience. I am not going to argue them out of it, and I am not going to pretend I hold it lightly.
I will not generalize any of this further until I see whether the people reading it stay the senior partner — whether the practice makes the questions harder to fool. That is the measurement — not whether you were ever fooled, but whether the practice keeps you asking. Send me the paragraph.