Fundamentally if I could have got the same outcome from an LLM as the “author” of an AI-generated piece, I value their effort at zero.

The more generative AI content there is in a piece, the less value it has per word.

Deep down our brains seem to know this, too.

LLM content is smooth, without novelty or surprises or shared amusement, and in text we have not summoned up ourselves, we are looking for that as a proxy for connection with the author.

I have used this analogy before but it’s like a rollercoaster without jerk or jounce; there are no thrills or interesting quirks for our brains.

So much so that when you see an AI-assisted piece of any length that has human-written sentences in it, they tend to stick out more, even before you have fully read the text, whereas the LLM words scan past so easily.

The thing that brings this home for me is that the final output of an LLM very rarely makes me chuckle at whimsy or weirdness, but the more disordered thinking traces more often do.