{"id":"W4389362477","doi":"10.48550/arxiv.2312.00805","title":"Gender inference: can chatGPT outperform common commercial tools?","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Sports Analytics and Performance","field":"Economics, Econometrics and Finance","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"University of Toronto","keywords":"Computer science; Inference; Artificial intelligence; Data science; Set (abstract data type); Generative grammar; Machine learning","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008372199,0.002792977,0.001303393,0.005484729,0.00110196,0.003221747,0.003729475,0.002429507,0.01236384],"category_scores_gemma":[0.03123676,0.0006262894,0.001791374,0.002828695,0.0008554583,0.00551525,0.003878724,0.00220003,0.01052199],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001414907,"about_ca_system_score_gemma":0.001926092,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008463539,"about_ca_topic_score_gemma":0.01622275,"domain_scores_codex":[0.9946801,0.001975971,0.0003582465,0.001400982,0.001185409,0.0003992641],"domain_scores_gemma":[0.9858973,0.009858068,0.000627716,0.001877619,0.001136724,0.0006024748],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003905962,0.000818915,0.07396789,0.002466992,0.0007217646,0.0006165087,0.001494135,0.01745973,0.004361182,0.01125058,0.2789452,0.6039912],"study_design_scores_gemma":[0.001348326,0.001550861,0.0427235,0.001279171,0.0006114696,0.00192019,0.004539933,0.49444,0.02162326,0.07708202,0.3522296,0.0006516606],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"software","genre_gemma":"empirical","genre_scores_codex":[0.2254974,0.01601441,0.2360384,0.01027628,0.006534006,0.001216373,0.08841702,0.3715456,0.04446052],"genre_scores_gemma":[0.5320088,0.003194507,0.3204365,0.005403467,0.001251279,0.001009842,0.1139445,0.01028338,0.01246776],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01236384,"threshold_uncertainty_score":0.04427695,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2764190071359,"score_gpt":0.2171654692307627,"score_spread":0.05925353790513727,"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."}}