AIGT detection on HC3 Pruthi attack Overall (in-domain)
100Overall AccuracyRoBERTa
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| RoBERTa2024.06 | 100 | 78.5 | 21.5 | 14,996.4 | |
| DeBERTa2024.06 | 100 | 67 | 33 | 13,472.16 | |
| Flooding2024.06 | 100 | 84.25 | 15.75 | 15,182.26 | |
| RanMASK2024.06 | 100 | 83.25 | 16.75 | 15,400.4 | |
| RMLM2024.06 | 100 | 84.75 | 15.25 | 15,161.52 | |
| SCRN2024.06 | 100 | 91.25 | 8.75 | 15,331.52 | |
| SeqXGPT2024.06 | 99.75 | 50.5 | 49.37 | 10,057.62 | |
| BERT2024.06 | 99.75 | 59.5 | 40.35 | 13,481.42 | |
| RDrop2024.06 | 99.75 | 84.25 | 15.54 | 14,824.54 | |
| ChatGPT-Detector2024.06 | 99 | 68 | 31.31 | 14,077.72 | |
| Log-Likelihood2024.06 | 98 | 50.5 | 48.47 | 9,259.91 | |
| Log-Rank2024.06 | 97.75 | 50 | 48.85 | 9,820.96 | |
| GLTR2024.06 | 97 | 48 | 50.52 | 9,914.36 | |
| Entropy2024.06 | 90.5 | 45.25 | 50 | 8,503.35 |