The Least Interesting Collaborator

I was listening to an episode of The Vergecast the other day about AI detectors, specifically the one that's supposedly reliable. The pitch is that it can look at a piece of writing and tell you whether a machine had a hand in it. And my first thought wasn't the one I think I was supposed to have. It wasn't panic. It was more of a shrug.

Because if you ran this blog through it, it would almost certainly light up. There's AI in here. I'm not going to pretend otherwise, and honestly the people who do pretend exhaust me. The whole thing has the frantic energy of a group of people trying to find out who microwaved fish in the office kitchen. Someone did it, everyone knows someone did it, and the investigation matters a great deal more to the investigators than it does to me. It was me. I microwaved the fish. There's AI in here.

The one AI detector people actually trust.

AI text detectors have been notoriously unreliable, but that’s starting to change. This year, Pangram keeps coming up as the trusted source in identifying AI-written text. We sit down with Pangram CEO Max Spero to find out how the system was made, how much we should trust it, and where the line is between useful AI and AI slop. [Media: https://www.youtube.com/watch?v=pVHSvJnSUNg]

The VergeJacob Kastrenakes

There's an assumption baked into the whole panic: that AI writing means AI generated the whole thing, start to finish, some person typing one lazy sentence and publishing whatever came back. All or nothing. No middle ground. It's a debate conducted entirely in black and white by people who would tell you, in any other context, that nothing ever is. So let me tell you how this actually works, because the mystery is the least interesting part.

An idea shows up, almost always at a moment I can't do anything about it. Driving, mid-ride, half asleep, in the shower, which is apparently where my brain keeps its best material and worst timing. So I make a note. A bad note, usually, three words that will mean nothing to me in a week, but enough to hopefully drag the thought back out later.

When I come back to it, if the shape is already clear in my head, I write an outline in my own words. Where it starts, what it's actually about, the couple of things I want to land. That part is me at a keyboard doing the normal, unglamorous work of figuring out what I think. My website platform is full of these, half-formed drafts and three-word notes waiting to be teased out into something, a backlog of ideas I fully intend to get to and probably never fully will.

And then there's the other part, sometimes I have a thought I can feel but cannot say. It's right there, fully formed as a feeling and completely useless as a sentence, and I sit and stare at it and produce nothing but worse and worse attempts. That's where I bring in the AI. To help me untangle it. I'll describe what I'm circling, badly, the same way I describe things to my wife on our walks, and we worry at it together until the thing I actually meant finally shows up on the page. The idea was always mine. What I got help with was finding the words for it.

That's the whole distinction, and it's the one that matters. At least to me. There's a difference between "write me a sixteen hundred word post about attention and the news cycle" and "I've got this thought about attention and I can't get it out, here's the mess in my head, help me see it." The first hands you a thought you never had. The second hands you back your own thought, clarified. Often this takes a few passes, me circling, the words getting closer each time, until the thing that's been sloshing around in my head finally sits still on the page. I live entirely in the second, and I've got no interest in the first. Not because it's cheating. Because it's boring. If the thought isn't mine, why would I want to publish it?

And honestly, this is the whole reason I've been able to write more than a handful of things at all. The ideas were never the problem. The problem was always the next step, taking a pile of loose, half-formed thoughts and wrestling them into something with a shape, a beginning and a point and an order that makes sense. That's the part that used to stop me cold, and it's the part this makes easier. Not the thinking. The assembling. And if there's anything more American than outsourcing the assembly, I haven't found it.

So does it matter? Whether there's AI in here, whether a detector would flag it, whether some purist would like to revoke my little writing hobby on principle?

No. Not even slightly. And I fully realize that "I don't care" can come off as a shrug from someone who hasn't thought about it. I've thought about it. I've arrived, deliberately, at not caring.

This is my personal website. Nobody assigned it. Nobody's paying for it. There's no editor, no grade, no client waiting on a deliverable. It exists because I wanted a quiet corner of the internet that was mine, and the whole point of a corner that's yours is that you get to decide how you keep it. I'm not entering these into a contest. I'm not passing them off as handmade artisanal sentences forged in a cabin by candlelight. I'm a guy writing things down and using the tools on the desk to do it better.

We've done this whole dance before, too, which is what makes the panic feel so familiar. When I was coming up, Google was the suspicious tool. There was a real stretch where using it to solve a problem marked you as a lesser engineer, someone who didn't truly know the material. I sat in interviews where you were expected to reproduce, from memory, on a whiteboard, some algorithm you would never once write from memory on the actual job, because on the actual job you would do what every working engineer does, which is look it up. Some interviewers would watch you reach for a search bar like they'd caught you cheating on a test. Meanwhile the entire industry ran, then as now, on the ability to find the answer, not to have memorized it. The tool that was disqualifying in the interview was mandatory the moment you got the job.

That gap between how we pretend the work happens and how it actually happens has always been where this kind of hand-wringing lives. Google didn't make engineers worse. It made the ones who could ask the right question faster. This feels like the same movie with a new tool in the lead role, and I've seen how it ends.

So do I owe every post a little label announcing there's AI inside? On my own website? The people who actually read this, who know my story or have read me long enough, can already tell these things are real. That was never something a disclaimer could prove anyway.