AIAI-writing detector reportedly reaches 98% accuracy on unseen blog posts
A user who helped train the model says it classifies blog posts by structural features, making 19 mistakes across 1,740 posts it had never seen.
TLDR
A user reports training a model to distinguish AI-written blog posts from human ones using structural features. They claim 98% accuracy on previously unseen posts, with 19 of 1,740 classified incorrectly, and share a link to a paper and code.
AI-writing detector reportedly reaches 98% accuracy on unseen blog posts
A user who helped train the model says it classifies blog posts by structural features, making 19 mistakes across 1,740 posts it had never seen.
