In this video, I break down why AI detection tools are unreliable (OpenAI shut down their own classifier after it performed worse than a coin toss), why gut-feeling prose analysis is even worse, why “certified human” badges are just an honor system that hurts the writers who need protection most, and why the asymmetry between a false accusation and a book sitting on a shelf should trouble all of us.

Another banger from Joan Westenberg.

You can’t reliably tell whether something was written by AI just by reading it. Or even running it through an AI-powered AI writing detection tool. Especially these dumb accusations about sentence structure, like “it’s not X, it’s Y.” That’s literally just language.

If I’m talking about something people suspect is X but it’s actually Y, what would you have me say? I don’t want to perform mental gymnastics and disrupt my natural writing flow to appease people who are paranoid they might read something an LLM generated.

And same for em dashes. Before it became a “telltale sign of AI,” I always used a double hyphen -- like this. Some writing platforms would automatically turn it into an en dash, but most of the time it was just two hyphens. I actually didn’t even realize you could type an em dash on an iPhone keyboard until this became such a hot-button topic — but now I use it all the time.