i’ve been thinking about whether the gap is closing or whether we’re just getting better at different things.
when AI writing tools first became widely used, the gap was obvious. the output was smooth in a suspicious way, the vocabulary was slightly formal, the transitions were textbook-perfect. anyone who wrote for a living spotted it immediately.
two years later, the surface-level tells are less consistent. the models have gotten better at sentence variety, at colloquial register, at avoiding the most obvious filler phrases. i’ve read AI-assisted pieces recently that i wouldn’t have flagged on instinct.
but. there’s a different gap that hasn’t closed and i’m not sure it’s closeable with the current approach. it’s not a surface feature – it’s depth. the difference between text that describes an idea and text that thinks through one. AI generates plausible content in the shape of thinking. it doesn’t actually work through a problem the way a writer does when they’re figuring something out in the process of writing.
that gap shows up most clearly in the places where the argument should get complicated – where a writer who’s really engaged would slow down, add a qualification, acknowledge something that cuts against their point. AI tends to keep moving. it doesn’t linger in the uncomfortable place the way real thinking does.
is that gap something that better models will close eventually? or is it something about the nature of what writing is for?
2025 or 2026?
The “thinking through versus describing” distinction is the most useful framing of this I’ve encountered. I’d push back slightly on whether it’s uncloseable – I think better models will get better at simulating the appearance of thinking-through, which is not the same as actually doing it, but may be indistinguishable at the surface level.
The honest answer is I don’t know where the ceiling is. What I do know is that the gap is real and mostly lives in exactly the places you’ve identified: the qualification, the uncomfortable acknowledgment, the moment where the argument earns its conclusion by surviving the hardest objection to it.
hot take: the gap isn’t really about model capability. it’s about what writing is supposed to do. if writing is a tool for communicating a conclusion, AI can do that. if writing is a process of working out what you think, then AI can’t do that by definition because there’s no thinking to work out.
the pieces where the gap is most visible are the ones where the writer is genuinely uncertain and that uncertainty is in the prose. AI doesn’t do genuine uncertainty. it does performed uncertainty, which reads differently.
to be fair, a lot of published human writing also doesn’t linger in the uncomfortable place. plenty of content writing, marketing copy, even a lot of journalism is essentially AI-shaped before AI existed – smooth, efficient, conclusion-forward, no real uncertainty in evidence. the comparison should probably be between AI and the best human writing, not AI and the median.
by that comparison the gap is still real and significant. but it’s also narrower than the discourse sometimes implies.
the performed uncertainty vs genuine uncertainty distinction is something i want to think more about. you can prompt a model to hedge, to acknowledge counterarguments, to qualify claims. the output looks like genuine uncertainty. but there’s something different about a writer who’s actually not sure and a model that’s been instructed to sound not sure.
honestly i’m not sure how to operationalize that distinction in a way that would survive empirical testing. which might be the point.
The question of whether better models will eventually close this gap is one I find genuinely uncertain. The capability trajectory suggests the surface features will keep improving. Whether the deep structural difference – text produced by thinking versus text produced by prediction – produces a gap that remains detectable is something I don’t think anyone knows yet.
What I’m more confident about is that the people best positioned to notice the gap as it narrows are the ones who read carefully and write seriously. Which is a reason to keep doing both.