RAID: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors

Benchmark Model Rank Results
ai-generated-text-detection-on-raidOriginality.ai#2Accuracy (FPR=5%): 85.0
ai-generated-text-detection-on-raidBinoculars#3Accuracy (FPR=5%): 79.6
ai-generated-text-detection-on-raidFast-DetectGPT#4Accuracy (FPR=5%): 73.6
ai-generated-text-detection-on-raidWinston AI#5Accuracy (FPR=5%): 71.0
ai-generated-text-detection-on-raidRADAR#6Accuracy (FPR=5%): 70.9
ai-generated-text-detection-on-raidGPTZero#7Accuracy (FPR=5%): 66.5
ai-generated-text-detection-on-raidGLTR#8Accuracy (FPR=5%): 62.6
ai-generated-text-detection-on-raidRoBERTa-Base (GPT-2 detector)#9Accuracy (FPR=5%): 59.1
ai-generated-text-detection-on-raidRoBERTa-Large (GPT-2 detector)#10Accuracy (FPR=5%): 56.7
ai-generated-text-detection-on-raidRoBERTa-Base (ChatGPT detector)#11Accuracy (FPR=5%): 44.8
ai-generated-text-detection-on-raidLLMDet#12Accuracy (FPR=5%): 35.0