DetectRL
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
DetectRL Multi-Domain
96.19AUROC
27
DetectRL Multi-LLM
97.17AUROC
27
DetectRL Prompt attack
80.9AUC@1%
20
DetectRL Out-of-Domain Multi-Topic 1.0 (test)
91.1Average Detection Score
18
DetectRL Mixing Attack
99.83AUROC (%)
16
DetectRL Perturbation Attack
99.99AUROC
16
DetectRL Paraphrase Attack
99.98AUROC
16
DetectRL Prompt Attack
99.96AUROC
16
DetectRL Direct Attack
99.99AUROC
16
DetectRL Llama-2
99.99AUROC
16
DetectRL PaLM-2
99.94AUROC
16
DetectRL Claude
1AUROC
16
DetectRL GPT-3.5
99.99AUROC
16
DetectRL Out-of-Domain Multi-LLM 1.0 (test)
90.6Average Performance Score
16
DetectRL Multi-Attack
99.56F1 Score (Prompt)
15
DetectRL Review
99.97AUROC
15
DetectRL Writing
99.99AUROC
15
DetectRL XSum
99.99AUROC
15
DetectRL ArXiv
100AUROC
15
DetectRL Multi-LLM (in-domain)
99.7Score (GPT-3.5)
14
DetectRL Multi-Topic (in-domain)
1arXiv Score
14
DetectRL
97.97AUROC (Multi-Domain)
12
DetectRL
97.2Accuracy
12
DetectRL Training Text: Llama-2-70b (test)
90.2Detection Score (Llama-2-70b)
12
DetectRL (test)
50.56Detection Score (Llama-2-70b)
12