I've started to notice that not everyone can tell when something is written in a typical "LLM voice" and I wonder if it's a new Dunning Kruger style signal that has quietly crept in without anyone giving a name to it yet.
I've seen a few instances recently where Twitter or YouTube users that I know aren't bots per se, are posting tweets or reading scripts that are obviously AI generated and then lying about it. I've been thinking about it and I've come to the conclusion that the only way a reasonable person would find that kind of lie defensible is if they truly can't see what is obvious to everyone else.
I'm sure there are people doing studies on this already so I should go look it up, but I wonder what the percentage of folks is right now who can't tell in a paragraph or two that a human didn't write something? I don't mean people who can't tell most of the time, I mean the people who can read an LLM generated book and upon completion hand it back saying, "It doesn't look like anything to me", West World style.
I'm probably making a bigger deal out of something that is likely to be a transient problem, someone will eventually figure out how to make LLMs sound more generic.
But what if we can't figure out how to cross that uncanny valley? Will we end up with a subset of people who all use "AI voice", essentially meat proxies for ChatGpt?
Something to think about, this probably should have been a Tweet but Elon ruined Twitter so you get this, cheers.