Writing and Sustainable Human/AI Collaboration

Writing is a genuinely important cognitive training — fiction, reports, essays, poetry, anything where you put real weight on output quality and quantity. The point of writing isn't cognitive training, of course; it's a medium for persisting thought and making it public. The training is just a means to that end. But now, after LLMs have taken over too much of our writing tasks, the training — or even just the experience — that writing gives you has become more important, and more uniquely valuable, than ever.

In undergrad, writing was instinct. Especially literary writing. I could sit down and bang out a 10-20k-character short story in an afternoon, no pauses. Poetic images spilled out of everything — objects, people, emotion — crossing time and physical constraints, connecting, shifting. Maybe half the time there was a poem in my head I could write, and I just couldn't be bothered to type it; only the very best ones got captured.

Essays too — my own love life, basketball write-ups, flat-out complaints — whatever I wanted to say, the words showed up on their own, unobstructed. No thinking about how the next sentence should be designed after this one, whether this paragraph was drifting off-topic, whether I could phrase something better. None of that. Always in one breath. A little smug to say, but it felt effortless. Ha.

Over the last year and change, my writing frequency dropped, and dropped, and dropped — a tangled set of reasons, but the endpoint was: basically never. And most of my work writing got covered by LLMs. My "writing activity" became one or two prompts to kick off an AI interaction. Sometime in January I sat down to write a basketball piece on Devin Booker. The first two or three paragraphs — I rewrote them over and over, none of them looked right. The mental friction was like a guy with 800-degree myopia climbing Mount Hua without his glasses: every step, look left, look right, feel the rock, pat the grass, then maybe continue. On one hand I was lamenting — I used to be a little soldier in the world of letters; I used to daydream about becoming a somewhat-serious writer; and now in my twenties I'm illiterate again, below elementary-school level. On the other hand I realized: if this keeps going, my brain is cooked. By over-outsourcing writing to LLMs, I'd accidentally outsourced a lot of the thinking itself. Brains are use-it-or-lose-it. Some abilities I'm fine losing — cooking, jump-shooting, fine. But writing, or rather the deep thinking, self-expression, synthesis, and critical reflection that sits behind writing — those I cannot afford to lose. If using LLMs ends with me getting dumber, I'd rather not use them.

These past two months I've forced myself to write, and at the same time I've been strict about monitoring my own AI interactions — making sure I don't skip any chance to train my brain. In practice that means: AI can do the work, but if it's new work, I have to at least learn how to do it first, and then have AI do it for me. I call this: sustainable human/AI collaboration.

Writing, I think, is the most important piece of that sustainable collaboration. If you ask me what the second piece is — reading. I was going to say "learning itself," but there are many ways to learn, and reading is the one that upgrades your language ability while also giving you knowledge. Wait — isn't this exactly like a large model? On one side we train the model toward better output (writing); on the other side we pipe it input (data). And reading is probably the one input type that most directly lifts general output quality. A human really is just an LLM.

I'll stop here — I need to go back to reviewing CSE 200 Complexity. No exam. Just review. If you've taken the class you get it: every lecture demands a review after.