Originality

Every generative model in production today is, by construction, a sophisticated average. It samples from what humans have already made. When it produces something that feels novel, the novelty is interpolative — a plausible combination of patterns it's already seen. This is not a flaw. It is the design.

What it means is that the model is structurally incapable of being the first one to make a particular thing. The first sentence of a genre. The first proof of a theorem. The first photograph that taught us to see a city differently. The thing that hadn't yet been made, and after which the average shifted to include it.

Originality is the bet you make against the average. It's the painter who spends fifteen years on a way of seeing that nobody asked for, and which later turns out to have been a door. It's the engineer who refuses the obvious architecture because she's chasing a smaller, weirder one she can't yet justify. It's the writer who keeps a sentence everyone tells her to cut because she knows it's doing the load-bearing work the rest of the paragraph won't.

The reason this matters now is that the cost of producing the average has gone to zero. Anybody can generate the median essay, the median image, the median pitch deck. The median is now noise. So the work that has any chance of mattering is the work that refuses to be the median — and deliberately so.

That's a strange place to live. It costs more. It looks wrong from the outside more often than not. It is, by definition, the thing nobody is asking for, until they are. And it is the one kind of work the AI tools genuinely cannot help you do, because if it could, it would already have done it, and then it would be average.

If you are the person in your field who keeps making the thing the model can't predict — even when it's slower, even when it costs you a paycheck — keep going. You are not redundant. You are the function that makes the rest of the system worth running.