매일 아침, 어제의 AI를 한 통으로 정리해 보내드립니다메일로 받아보기

METAL LAB

MD-ProTector: Positioning Multiple Data-Driven Prototypes for LLM-Generated Text Detection

arXiv:2608.104592026-08-12

arXiv:2608.10459v1 Announce Type: new Abstract: As LLM-generated content becomes more sophisticated, detection systems for distinguishing those texts from human-written text must operate at scale while handling diverse writing styles, domains, languages, and generator models. Input-only encoder detectors are suitable for practical deployment setting, but standard binary classification supplies only the class label and does not explicitly organize the substantial variation within either class. We

저자 · Jinmo Han, Jimin Hong, Chanyeong Moon, Ju Yeon Kang, Seonuk Kim, Nam Soo Kim

arXiv에서 원문 보기