Thank you for taking a look! Yeah there's definitely a few tells which are noticable if you look for it :)

Thank you for trying! I first built it as 'detect the human' response, but that was counter to the 'slop or not' framing. Yeah I'm also observing the same based on the first few hundred people's results. The harder models seem to write almost too well and that's generally not how humans write on the internet unless it is a blog post/essay. The easier models seem to be the ones that are tripping people the most.

Ha :) I'm not building models, nor am I affiliated with any big labs. The idea is to use this to educate people how to spot tells of AI writing. Although like any data that's made open this can be used to train future models as well I suppose.

thank you so much for taking a look :) Yeah you'd be surprised how difficult it can get to spot nuances sometimes. Sometimes, there isn't any nuance at all and the AI is just as good at writing about something pretending to know about the topic.

You'd be surprised at the nuances we tend to miss :)

This time around I prompted the models not necessarily to be adversarial - i didn't ask them to try and fool the reader. But i gave them contextual info - something to the effect of "you're a user posting on hacker news"