1 paper · 1 filter
Wan Tian, Zhongyi Li, Yawen Li +3
Mixed human-LLM documents require locating authorship transitions from detector scores whose reliability varies across text units. Existing weighted mean contrasts are vulnerable t…