An algorithm to detect an algorithm

Someone ran one of my posts through an AI detector this week. It came back machine-written. Which is funny, because the detector is an algorithm too, guessing whether another algorithm wrote your words. It takes one to know one.

An AI detector flagged one of my posts as machine-written this week. Fair guess. It was a post about AI, written by a guy who builds AI for a living.

But here’s the joke. The detector is the same kind of thing it’s flagging: a transformer trained partly on GPT-4. An algorithm built to hunt algorithms.

And these things barely work. OpenAI built one, then killed it in 2023 because it caught almost nothing and kept flagging real people. The company that makes the models couldn’t spot the models. I’m not going to sweat a score from a smaller one.

The score doesn’t interest me. How I work does. I use AI in the open, so here’s what that actually looks like.

What I actually do

I draft with AI. I use Grammarly. Whatever gets an idea out of my head and into a draft.

Then I rewrite it until it sounds like me. Shorter. Blunter. With the examples and opinions a model doesn’t have.

The tool greases the wheels. The judgment and the voice are mine. That’s the part no tool does for you.

The human behind it

I’ve got a full-time job. I coach two hockey teams. Some months that’s a hundred-plus hours away from my desk.

I coach for one reason. I want these kids to love the game I grew up on, and to have someone in their corner while they figure it out. Best case, a few of them go on to coach their own teams one day and do it better than me. That’s how you give it back.

So no, I’m not writing these at a desk with a spare afternoon. AI is how the idea gets from my head to the page in the margins of a day like that. Not instead of me.

The experience is mine. The opinion is mine. The time is the part I don’t have. That’s what the tool is for.

Where the ideas come from

The ideas don’t come from a model making things up. They come from my own second brain.

I keep a vault. Notes, a running journal. I see something worth keeping during a day of building, it goes in. My own agents feed it too, daily and weekly, so it fills up while I work.

So I don’t sit down and ask a model what to think about AI. I pull from a pile of things I actually ran into, doing the job. The well doesn’t run dry.

The only test that matters

I don’t run my posts through a detector. I don’t rewrite them to beat one.

I ask two questions. Is it true? Does it sound like me?

If both are yes, I post it. A detector can call it a robot all it likes. I know who wrote it, and so does anyone who’s read more than one thing I’ve put out.

The goal was never to beat an algorithm. It was to say something worth reading, in my own voice. No tool can fake that, and no tool can flag it away.