LLM-generated text detection on KatFish Essay
97.57AUC-ROC (Solar)KatFishNet
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| KatFishNetFeature=Punctuation2025.02 | 97.57 | 94.63 | 92.45 | 94.88 | |
| KatFishNetFeature=POS Combinations2025.02 | 92.26 | 83.1 | 73.63 | 82.99 | |
| Exaone 3.5Category=LLM Paraphrasing2025.02 | 92.08 | 79.74 | 72 | 81.27 | |
| KatFishNetFeature=Word Spacing2025.02 | 86 | 80.63 | 71.91 | 79.51 | |
| Log-LikelihoodCategory=Confidence2025.02 | 83.84 | 23.89 | 66.2 | 57.97 | |
| Log-RankCategory=Confidence2025.02 | 78.84 | 20.66 | 61.92 | 53.8 | |
| RoBERTaCategory=Fine-tuning2025.02 | 66.77 | 66.65 | 64.37 | 65.93 | |
| NPRCategory=Perturbation2025.02 | 55.22 | 19.9 | 44.71 | 39.94 | |
| DetectGPTCategory=Perturbation2025.02 | 52.78 | 37.45 | 47.18 | 45.8 | |
| Exaone 3.5Category=LLM Prompting2025.02 | 50.42 | 49.74 | 50.07 | 50.07 | |
| LRRCategory=Confidence2025.02 | 45.08 | 80.56 | 53.15 | 59.59 | |
| EntropyCategory=Confidence2025.02 | 31.25 | 84.53 | 44.12 | 53.3 |