Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

HC3

Benchmarks

Task NameDataset NameSOTA ResultTrend
AIGT detectionHC3 PWWS attack, AI to Human (in-domain)
Overall Accuracy100
28
AI Text DetectionHC3 AI-Prob
Finance Accuracy84.1
24
AI-generated text detectionHC3 (test)
F1 (Overall)99.93
18
AIGT detectionHC3 Pruthi attack Overall (in-domain)
Overall Accuracy100
14
AIGT detectionHC3 Pruthi attack Human to AI (in-domain)
OA1
14
AIGT detectionHC3 Pruthi attack AI to Human (in-domain)
Overall Accuracy100
14
AIGT detectionHC3 Deep-Word-Bug attack Overall (in-domain)
OA100
14
AIGT detectionHC3 Deep-Word-Bug attack Human to AI (in-domain)
Overall Accuracy100
14
AIGT detectionHC3 Deep-Word-Bug attack AI to Human (in-domain)
Overall Accuracy100
14
AIGT detectionHC3 PWWS attack, Human to AI (in-domain)
OA100
14
Machine-Generated Text DetectionHC3 COLING2025 (val)
AUC99.99
13
AI-text detectionHC3
AUROC100
10
Machine-Generated Text DetectionHC3 (full)
Accuracy99.92
9
LLM-generated text detectionHC3 Chi-Psy
TPR @ FPR=1%90
3
Machine-generated text detectionHC3 Cross-domain
Score (Medicine)80.48
3
AI-Text DetectionHC3 PLUS (test_si)
Balanced Accuracy (BA)87.21
2
AI-Text DetectionHC3 PLUS QA (test)
Balanced Accuracy99.47
2
AI-Text DetectionHC3 SI (Single-Iteration)
Balanced Accuracy86.85
1
AI-Text DetectionHC3 QA
Balanced Accuracy99.63
1
Showing 19 of 19 rows