Tag: training data

A new arXiv paper labels 31.1% of filtered August web tokens as AI-written — then shows, across 800 models, when those tokens stop helping.