AI-generated text detection on M4
93AUROCLUSR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LUSRSupport size (k)=52026.06 | 93 | — | |
| Exons-DetectTraining paradigm=Training-free2026.03 | 92.43 | 88.05 | |
| DNA-DetectLLMTraining paradigm=Training-free2026.03 | 91.74 | 87.72 | |
| Lastde++Training paradigm=Training-free2026.03 | 91.43 | 84.97 | |
| BinocularsTraining paradigm=Training-free2026.03 | 90 | 87.4 | |
| Fast-DetectGPTTraining paradigm=Training-free2026.03 | 89.77 | 84.12 | |
| LAPDStandardization=true, Base Model=Llama2-7B, Aligned Model=Llama2-7B-Instruct2026.04 | 88.02 | — | |
| LAPD2026.04 | 88.02 | — | |
| LogRankTraining paradigm=Training-free2026.03 | 87.5 | 80.7 | |
| LikelihoodTraining paradigm=Training-free2026.03 | 85.77 | 78.38 | |
| EntropyTraining paradigm=Training-free2026.03 | 83.72 | 79.1 | |
| BinocularsStandardization=false2026.04 | 83.42 | — | |
| LUAR MultidomainSupport size (k)=52026.06 | 83 | — | |
| LUAR CRUDSupport size (k)=52026.06 | 82 | — | |
| LUAR Multi-LLMSupport size (k)=52026.06 | 82 | — | |
| Fast-DetectGPTStandardization=true2026.04 | 81.1 | — | |
| CISRSupport size (k)=52026.06 | 81 | — | |
| BiscopeTraining paradigm=Training-based2026.03 | 79.74 | 73.08 | |
| Lastde++Standardization=true2026.04 | 79.44 | — | |
| LogRankStandardization=false2026.04 | 78.69 | — | |
| LikelihoodStandardization=false2026.04 | 77.63 | — | |
| OpenAI-DTraining paradigm=Training-based2026.03 | 77.51 | 71.18 | |
| DNA-DetectLLMStandardization=true2026.04 | 77.38 | — | |
| ReMoDetect2026.04 | 74.98 | — | |
| ImBD2026.04 | 73.41 | — | |
| DetectGPTTraining paradigm=Training-free2026.03 | 73.13 | 70.11 | |
| OpenAI Detector2026.04 | 72.24 | — | |
| IRMTraining paradigm=Training-free2026.03 | 71.85 | 70.75 | |
| EntropyStandardization=false2026.04 | 71.24 | — | |
| RAIStandardization=false, Base Model=Llama2-7B, Aligned Model=Llama2-7B-Instruct2026.04 | 70.46 | — | |
| LUSRSupport size (k)=12026.06 | 67 | — | |
| ProtoNetSupport size (k)=52026.06 | 64 | — | |
| R-DetectTraining paradigm=Training-based2026.03 | 61.91 | 67.14 | |
| LUAR CRUDSupport size (k)=12026.06 | 60 | — | |
| LUAR MultidomainSupport size (k)=12026.06 | 60 | — | |
| LUAR Multi-LLMSupport size (k)=12026.06 | 58 | — | |
| CISRSupport size (k)=12026.06 | 58 | — | |
| SBERTSupport size (k)=52026.06 | 57 | — | |
| ProtoNetSupport size (k)=12026.06 | 55 | — | |
| SBERTSupport size (k)=12026.06 | 50 | — | |
| DetectGPTStandardization=true2026.04 | 46.54 | — |