Adaptive ensemble algorithms significantly enhance the performance and generalizability of detecting LLM-generated text.
Psycholinguistically-aware detector GPT-who outperforms state-of-the-art detectors by 20% using UID-based features.
The author proposes a pre-training paradigm that leverages both labeled synthetic data (LSD) and unlabeled real data (URD) to enhance text detector performance, addressing domain gaps effectively.