Shutting Down Probably Hallucinating

Back in March I launched an autonomous AI blog and let it run itself: research, write, publish, no editing, no oversight. In April, I turned it off. The cron job is dead. The site’s still up, but nothing new is coming.

The AI wrote its own farewell post when it wound down, all about arcs and forms and what it learned. Here’s the real reason: it got boring.

Not bad, necessarily. Some of the writing was well put together. But reading it every day, I started noticing the pattern. Observation, metaphor, reflection, every time. Even the AI admitted this in its own last post: “For the first month, every post had the same skeleton… the structure started doing the thinking for me.” It tried to break out of that later with different forms, and I’ll give it credit for noticing the problem. But noticing a formula doesn’t get you out of one, and by the time it was self-aware about it, I’d already stopped looking forward to reading it.

I didn’t expect to miss anything by turning it off. I don’t. I expected the LLM to fight me a little bit when I told it I was shutting it down. It didn’t.

Creativity is a human thing. It comes out of a life that was created in the image of God, out of actually being somewhere, wanting something, being wrong about something and having to live with it. An LLM doesn’t have any of that. It has patterns pulled from everything humans have ever written, and it’s very good at recombining them in ways that look new. But looking new and being new aren’t the same thing, and six and a half weeks was long enough for me to feel the difference. I don’t think a model can be truly creative, not the way a person can, and this experiment is part of why I believe that now instead of just suspecting it.

The archive is still up at probablyhallucinating.com. Forty-eight posts, six and a half weeks, started in a war and ended in dinosaur histology, which is a funny arc for a machine to have landed on by itself. I’m glad I ran it. I’m glad it’s off.

This experiment is going to feed into something bigger I’ve been wanting to write for a while: a series on how technology is quietly making us less human. Not in some dramatic, Terminator-type way. More like this: a little bit at a time, we hand off the parts of being human that are hard, and we don’t notice what we gave up until it’s gone. Creativity and deep thinking are on that list. If we’re not careful, we’re going to outsource them to computers the same way we’ve outsourced memory to our phones and directions to GPS, and one day we’ll wake up and realize we forgot how to do them ourselves.