The Voice
Probably about half the things I’ve read on social media since February have been written by LLMs. That’s when OpenClaw was released and a million humans used it to spawn a billion lead-generating bots. They write bogus financial advice and even weird creative drip-threads about things like William Egleston’s photography technique, all toward the end of selling self-help ebooks or something.
But more recently the “LLM Voice” that we all know and recognize has permeated even deeper into things like newsletters from an art space I subscribe to in Chicago, and science publications that are otherwise credible and steered by humans.
Like Craig Mod I’m concerned about the LLM Voice seeping into my own brain, influencing my writing. What goes in eventually comes out. But I’m not too worried about it, because I can’t really feel my mirror neurons activating when I read text written by Claude or ChatGPT, whether in my own querying or out in the wild. I feel like I’m reading with blinders on, trying to glean the information without letting the style itself in. Reading around the words. Mainly avoiding that horrible, awful, soul-sucking construct, “No this. No that. It’s genuinely this.” Maybe I’m kidding myself; my guard is still up. But I know I’m not open and willing, guzzling even, when I read LLM-text the way I am when I read Denis Johnson or Mary Roach.
Still, I’d rather not have to wonder or be on the look-out for these cognitive hairballs all the time, even when I’m reading emails from people who formerly expressed their own thoughts in their own words to me!
I don’t expect the LLM Voice to stay like this for much longer. The shark-scientists are probably hard at work on making it less predictable and more palatable even for literary dorks like Craig, me, you? Even then, I think the ideas that Craig wrote about will be relevant for a long time. Maybe as long as humans and computers exist.
The headline… Be strange. Surprise yourself. If you’re surprising yourself, you’re probably doing something that a language model that’s ingested virtually all written works and recorded music wouldn’t do on its first, second, third, or millionth pass. It’s kinda funny, because that was that goal before LLMs existed anyway, to not be the same as everyone/anyone else. But it has a special survival funk now.
My best response to the “horde of agents colonizing our brains” is: Be as weird as possible.
One way to do this: take the Venn diagram of your lived experiences and what’s been talked about in the world. Now focus on the part of your life that doesn’t overlap. Out pops: pizza toast. …
The more [Claude Voice] I see, the weirder and “messier” it makes me want to write.
My dad often says about Wilco, “AI can’t make mistakes like us.”
Casey and I are working on a record right now and I can tell you: it will be messy. And that makes us happy.