List your company

Machine-Text Detection on Blogs (evaluation)

1AUROCBaseline
0.5740160.6846080.79520.905792May 20, 2025

Evaluation results

MethodLinks
BaselineSample Size (N)=10, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.051
BaselineSample Size (N)=25, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.051
BaselineSample Size (N)=50, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.051
Style-aware ParaphraserSample Size (N)=25, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.051
Style-aware ParaphraserSample Size (N)=50, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.051
BaselineSample Size (N)=5, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.050.9998
Style-aware ParaphraserSample Size (N)=10, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.050.9764
BaselineSample Size (N)=1, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.050.9074
Style-aware ParaphraserSample Size (N)=5, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.050.8675
Style-aware ParaphraserSample Size (N)=1, Target Models=Qwen3-8B, Qwen3-14B, and Mistral-Nemo2025.050.5904