{"id":"W4416394486","doi":"10.48550/arxiv.2510.07500","title":"Black-Box Detection of LLM-Generated Text Using Generalized Jensen-Shannon Divergence","year":2025,"lang":"en","type":"preprint","venue":"ArXiv.org","topic":"Authorship Attribution and Profiling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Government of Canada; Canadian Institute for Advanced Research","keywords":"Divergence (linguistics); Detector; Asymptotic distribution; Discretization; Pattern recognition (psychology); Matrix (chemical analysis); Statistical hypothesis testing","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0007119076,0.0004049576,0.0005457081,0.0003293384,0.0002624183,0.00009794241,0.001210533,0.0005197796,0.00005005037],"category_scores_gemma":[0.0001770643,0.0004238431,0.000254352,0.0009807465,0.0001126864,0.0002204026,0.001869076,0.0007028734,0.00006424967],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001886283,"about_ca_system_score_gemma":0.0003969747,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004716294,"about_ca_topic_score_gemma":0.00003349643,"domain_scores_codex":[0.9970033,0.0004380421,0.0007442965,0.0009730086,0.000396145,0.0004452371],"domain_scores_gemma":[0.9975415,0.00008503291,0.0005824405,0.001133709,0.0005255229,0.0001317912],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0001829798,0.000389718,0.1037648,0.001055767,0.0005284196,0.00006506741,0.003481654,0.079039,0.7876146,0.006640968,0.0003398946,0.01689717],"study_design_scores_gemma":[0.0004055196,0.00005489019,0.01119751,0.0003102747,0.00007503404,0.000006030607,0.00002875376,0.361257,0.6242516,0.001024576,0.0007749411,0.0006138468],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5583532,0.0003478454,0.4386412,0.0001311194,0.001949014,0.0002416123,0.00002701567,0.0001768416,0.0001321234],"genre_scores_gemma":[0.9818695,0.0001556512,0.01654304,0.0001955938,0.0001729288,0.00001799445,0.00004503367,0.00001721279,0.000983051],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4235163,"threshold_uncertainty_score":0.9998214,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09692107733937994,"score_gpt":0.3169443602513833,"score_spread":0.2200232829120033,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}