Machine-Generated Text Detection on Essay ChatGPT-turbo
95.78TPR@FPR=1%Likelihood-M
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Likelihood-MTraining Source=Claude, Detection Method Variant=Markov-informed score calibration2026.02 | 95.78 | 99.66 | |
| Log-Rank-MTraining Source=Claude, Detection Method Variant=Markov-informed score calibration2026.02 | 94.8 | 99.66 | |
| Log-RankTraining Source=Claude, Detection Method Variant=Standard2026.02 | 80.89 | 99.24 | |
| LikelihoodTraining Source=Claude, Detection Method Variant=Standard2026.02 | 73.6 | 99.13 | |
| Entropy-MTraining Source=Claude, Detection Method Variant=Markov-informed score calibration2026.02 | 39.69 | 96.91 | |
| DetectGPT-MTraining Source=Claude, Detection Method Variant=Markov-informed score calibration2026.02 | 29.11 | 80.33 | |
| FastGPTTraining Source=Claude, Detection Method Variant=Standard2026.02 | 16.89 | 88.62 | |
| DNA-GPT-MTraining Source=Claude, Detection Method Variant=Markov-informed score calibration2026.02 | 13.33 | 76.3 | |
| EntropyTraining Source=Claude, Detection Method Variant=Standard2026.02 | 13.24 | 95.94 | |
| DetectGPTTraining Source=Claude, Detection Method Variant=Standard2026.02 | 2.36 | 55.34 | |
| FastGPT-MTraining Source=Claude, Detection Method Variant=Markov-informed score calibration2026.02 | 0.84 | 54.76 | |
| DNA-GPTTraining Source=Claude, Detection Method Variant=Standard2026.02 | 0.53 | 40.65 |