AI-text detector evasion on M4 evaluation set
90MAGE ASRMAGE BERTScore Linear 0.1
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MAGE BERTScore Linear 0.1schedule=Linear 0.12026.05 | 90 | 79.1 | 21 | 86 | 6 | 39 | 13 | 97 | 15 | 95 | 15 | |
| MAGE BERTScore 0weight=02026.05 | 85 | 80.4 | 18 | 85 | 3 | 64 | 4 | 98 | 27 | 96 | 24 | |
| DEPO2026.05 | 70 | 85.4 | 4 | 83 | 2 | 51 | 4 | 95 | 22 | 93 | 17 | |
| MAGE BERTScore Linear 0.3schedule=Linear 0.32026.05 | 68 | 83.9 | 4 | 85 | 4 | 57 | 1 | 96 | 21 | 92 | 18 | |
| MAGE BERTScore Linear 0.5schedule=Linear 0.52026.05 | 29 | 82.7 | 23 | 73 | 9 | 18 | 29 | 97 | 3 | 87 | — | |
| Adv. Paraphrase2026.05 | 13 | 85.6 | 39 | 88 | 1 | 73 | 2 | 80 | 1 | 79 | 9 | |
| Direct Paraphrasing2026.05 | 6 | 85.6 | 41 | 77 | 4 | 38 | 13 | 93 | 1 | 77 | 7 | |
| AuthorMist2026.05 | 4 | 81.7 | 42 | 82 | 0 | 48 | 5 | 98 | 26 | 97 | 23 | |
| SilverSpeak2026.05 | 3 | 39.1 | 35 | 0 | 48 | 0 | 30 | 66 | 45 | 99 | 28 | |
| Original2026.05 | 2 | — | 47 | 84 | 3 | 36 | 14 | 93 | 8 | 74 | 1 | |
| MAGE BERTScore 1weight=12026.05 | 2 | 99 | 46 | 83 | 0 | 36 | 14 | 93 | 6 | 73 | 3 | |
| StealthRL2026.05 | 1 | 65.6 | 34 | 77 | 15 | 2 | 44 | 100 | 43 | 99 | 41 |