List your company
ResearchFormerly WizWand

LLM-generated text detection on Xsum, WritingPrompts, and SQuAD (GPT-5-Chat test)

81.33AUROCSurpMark
40.385251.015161.64572.2749Oct 8, 2025
Updated 5mo ago

Evaluation results

MethodLinks
SurpMarkBlack-box setting=true, k (number of discretization states)=7, B (reference-per-test budget)=22025.1081.33
SurpMarkBlack-box setting=true, k (number of discretization states)=6, B (reference-per-test budget)=22025.1078.33
R-DetectBlack-box setting=true2025.1067.75
FourierGPTBlack-box setting=true2025.1064.82
DetectNPRBlack-box setting=true, B (reference-per-test budget)=1002025.1054.99
DetectGPTBlack-box setting=true, B (reference-per-test budget)=1002025.1054.6
LogRankBlack-box setting=true2025.1049.83
DetectLRRBlack-box setting=true2025.1049.83
DNA-GPTBlack-box setting=true, B (reference-per-test budget)=102025.1049.82
BinocularsBlack-box setting=true2025.1049.65
LikelihoodBlack-box setting=true2025.1049.62
EntropyBlack-box setting=true2025.1046.99
Lastde++Black-box setting=true, B (reference-per-test budget)=1002025.1043.51
Fast-DetectGPTBlack-box setting=true2025.1043.39
LastdeBlack-box setting=true2025.1041.96