{"id":"W4409258158","doi":"10.1613/jair.1.16665","title":"Detecting AI-Generated Text: Factors Influencing Detectability with Current Methods","year":2025,"lang":"en","type":"article","venue":"Journal of Artificial Intelligence Research","topic":"Topic Modeling","field":"Computer Science","cited_by":40,"is_retracted":false,"has_abstract":true,"ca_institutions":"National Research Council Canada","funders":"Ministère de la Défense Nationale","keywords":"Current (fluid); Computer science; Artificial intelligence; Physics","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03233095,0.001458091,0.001389667,0.006893528,0.002117226,0.008502508,0.002819546,0.002971203,0.002855466],"category_scores_gemma":[0.2383678,0.0005881769,0.001186764,0.004880014,0.002266101,0.01475893,0.003054719,0.003825939,0.003835718],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001442375,"about_ca_system_score_gemma":0.001611154,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004329859,"about_ca_topic_score_gemma":0.004582524,"domain_scores_codex":[0.9726078,0.01420758,0.00305499,0.004326969,0.005052515,0.0007501737],"domain_scores_gemma":[0.6395614,0.307193,0.01313207,0.02609702,0.01219128,0.001825278],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001745343,0.0006932091,0.3704461,0.002270594,0.0007209558,0.000477067,0.004773808,0.0190265,0.007965226,0.01272688,0.04045288,0.5387014],"study_design_scores_gemma":[0.0002283044,0.0005967573,0.1382182,0.00128442,0.0006804547,0.002593728,0.006381493,0.6894081,0.02510892,0.07747814,0.05768181,0.0003395933],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5131224,0.03063308,0.3947145,0.01596719,0.001718416,0.00127631,0.009720131,0.007247591,0.02560031],"genre_scores_gemma":[0.899339,0.003534572,0.08482188,0.0009293738,0.001085131,0.0003628375,0.006872129,0.0005274373,0.00252753],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03233095,"threshold_uncertainty_score":0.1709844,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2192959255969502,"score_gpt":0.498691993893544,"score_spread":0.2793960682965937,"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."}}