{"id":"W6892830738","doi":"10.5281/zenodo.12011536","title":"Scarlet Cloak and the Forest Adventure: The Issue of False Positives in AI Detection Tools","year":2024,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Computational and Text Analysis Methods","field":"Social Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Lethbridge","funders":"","keywords":"Generative grammar; Grammar; False positive paradox; Production (economics); Selection (genetic algorithm); Discipline; Natural language generation; Prototype theory","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002179513,0.00005295286,0.00008423359,0.0001132061,0.001217827,0.0006356454,0.000324792,0.00002887698,0.000888997],"category_scores_gemma":[0.0008602458,0.00003443833,0.00004705979,0.0007551542,0.0004461508,0.000201092,0.000235293,0.00016343,0.0001771648],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003996075,"about_ca_system_score_gemma":0.000005499172,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000299316,"about_ca_topic_score_gemma":0.00005158916,"domain_scores_codex":[0.9982262,0.001074182,0.0001493021,0.0001565131,0.0002676694,0.0001261585],"domain_scores_gemma":[0.9993573,0.0002714107,0.00003909135,0.000107012,0.000189243,0.00003597553],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00008498917,0.00003632831,0.00001991133,0.00003390757,0.00006648371,0.000004412178,0.02032444,0.0002689616,0.0006140219,0.1881226,0.005646115,0.7847778],"study_design_scores_gemma":[0.0002523331,0.00004258598,0.004459384,0.00002774481,0.00002920702,0.000009173071,0.002990807,0.003174664,0.0001160431,0.01080428,0.9780278,0.00006601342],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4407246,0.009914187,0.08096796,0.1307157,0.001045781,0.003740286,0.0002514099,0.001323925,0.3313161],"genre_scores_gemma":[0.998848,0.0001982551,0.00004944541,0.0001378449,0.0001288899,6.559025e-8,0.00003220141,0.00009444206,0.0005108647],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9723817,"threshold_uncertainty_score":0.9733894,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02969739178697276,"score_gpt":0.3148960804134628,"score_spread":0.28519868862649,"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."}}