back
1 comments
Doesn't seem like it solves the diversity problem, it just makes writing seem more human-like by selected metrics. Different patterns but same annoyance if you spam them enough.

>a sample of 100 model outputs scored as 100% human written by Pangram AI detector

Of course it did. It's trained on default writing styles of major LLMs. You can cheat it by prompting them to write in another style, because "100% human" actually means "I can't detect anything I've been trained on".

>LLMs are not the cause of slop. Lack of effort/care is.

This is right. Nobody cares about creative writing, because everyone is chasing coding machines, robotics, and the AGI pie in the sky.