AI-Generated Text Detection on mixed-source AI -> Human GPT-2, GPT-Neo, GPT-J, LLaMa, GPT-3
96.5Overall AccuracySeqXGPT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SeqXGPTAttack=PWWS2024.06 | 96.5 | 65 | 32.64 | 1,867.81 | |
| RDropAttack=PWWS2024.06 | 96.5 | 69 | 28.5 | 1,819.95 | |
| SeqXGPTAttack=PWWS2024.06 | 96.25 | 67.5 | 29.87 | 1,880.9 | |
| FloodingAttack=PWWS2024.06 | 96 | 60.5 | 36.98 | 1,800.01 | |
| DeBERTaAttack=PWWS2024.06 | 95.75 | 67.25 | 29.77 | 1,852.7 | |
| FloodingAttack=PWWS2024.06 | 95.75 | 56.75 | 40.73 | 1,705.23 | |
| RDropAttack=PWWS2024.06 | 95.75 | 69.5 | 27.42 | 1,817.78 | |
| RoBERTaAttack=PWWS2024.06 | 95.5 | 64.5 | 32.46 | 1,840.19 | |
| DeBERTaAttack=PWWS2024.06 | 95.5 | 54.5 | 42.93 | 1,764.94 | |
| SCRNAttack=PWWS2024.06 | 95.5 | 89.25 | 6.54 | 2,043.44 | |
| SCRNAttack=PWWS2024.06 | 95 | 87 | 8.42 | 1,986.98 | |
| RoBERTaAttack=PWWS2024.06 | 94.25 | 63.5 | 32.63 | 1,784.96 | |
| RanMASKAttack=PWWS2024.06 | 94 | 60 | 36.17 | 1,784.11 | |
| RMLMAttack=PWWS2024.06 | 91.25 | 73.5 | 19.45 | 1,933.23 | |
| RMLMAttack=PWWS2024.06 | 91 | 69 | 24.18 | 1,879.96 | |
| BERTAttack=PWWS2024.06 | 90.5 | 1 | 98.9 | 1,204.52 | |
| BERTAttack=PWWS2024.06 | 90.25 | 30 | 66.76 | 1,509.98 | |
| RanMASKAttack=PWWS2024.06 | 90 | 65.5 | 27.22 | 1,749.92 | |
| GLTRAttack=PWWS2024.06 | 76.5 | 0 | 100 | 1,260.99 | |
| Log-RankAttack=PWWS2024.06 | 73.5 | 0.5 | 99.32 | 1,286.24 | |
| Log-LikelihoodAttack=PWWS2024.06 | 72 | 0.5 | 99.31 | 1,281.86 | |
| GLTRAttack=PWWS2024.06 | 72 | 9.5 | 86.81 | 1,273.32 | |
| Log-RankAttack=PWWS2024.06 | 68 | 28.25 | 58.46 | 1,491.72 | |
| Log-LikelihoodAttack=PWWS2024.06 | 67 | 27 | 59.7 | 1,474.88 | |
| EntropyAttack=PWWS2024.06 | 63 | 0 | 100 | 1,239.29 | |
| EntropyAttack=PWWS2024.06 | 59.25 | 13.75 | 76.79 | 1,317.84 |