They’re using a crude linear model to identify AI output, which is not that much different from that which the AI safety industry is peddling, or the people that sell solutions to identify AI output.

You can’t reliably predict the output of a non-linear model with a simple linear model, no matter how hard you wish it.

Perfectly said. The safety industry is such a grift, propped up by ignorant VCs who don't want to miss their rocket ship

That’s not an LLM indicator —it’s totally acceptable grammar.