Concetti Chiave
This research paper introduces and evaluates new approaches for accurately identifying machine-generated text segments within partially machine-generated documents, demonstrating significant improvements over existing methods and highlighting potential applications for detecting AI-generated content.
Statistiche
The DeBERTa-CRF model achieved an MAE of 18.538 on the test set, which consists of unseen domains and generators.
The Longformer.pos-CRF model achieved an MAE of 18.542 on the test set.
ZeroGPT, a proprietary system, achieved an average sentence accuracy of 0.7976 on the development set.
The proposed model achieved an average sentence accuracy of 0.9848 on the development set and 0.9974 on the test set.