Post

What's depressing, and what the study reveals, is that the vast majority of readers aren't paying much attention & don't have much discernment.

I'm old enough to remember that that's what it sounded like when the record reached its end, and the needle is just going around on the innermost groove (and there's often a specific crackle each time the needle reaches the point where the spiral groove reaches that innermost circle.

Well, given that AI draws entirely on human-created content, I guess it stands to reason.

I'd be curious about a follow up study where the population is specific. Software engineers spend a lot of time engaging with AI authored text now and I can feel a distinct difference when it's used. But I don't know if that transfers generally to all types of writing.

That is not what slush pile readers say (Talk to Neil Clarke, say). I suspect the key ingredient is 'practise'

And that's assuming the "humans" didn't in fact use AI to "help" with their stories.

My guess would be that the quality of AI writing would be narrower. Like AI isn't going to write something as evocative as Ursula K Le Guin, but it also is now good enough to not sound like someone's teenage fanfic where they accidentally changed a character's name in the middle ...

Anecdotally, it seems the newer models are noticeably getting worse at generating words for humans - as they are being more tuned for “agentic” work and code. (And probably training on more AI output )

AI stories are “created” from scraped human writings. So this is no surprise. When AI checkers first came out, I ran 20 pre AI client articles I’d written through them. 18 came back as “likely AI.”