{"id":"W4362456272","doi":"10.1521/soco.2023.41.2.103","title":"Accuracy and Consistency in Social Categorization Across Context, Motivation, and Time","year":2023,"lang":"en","type":"article","venue":"Social Cognition","topic":"Evolutionary Psychology and Human Behavior","field":"Psychology","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Categorization; Psychology; Consistency (knowledge bases); Perception; Cognitive psychology; Social psychology; Social perception; Snapshot (computer storage); Artificial intelligence; Computer science","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":[0.0002645963,0.00009340775,0.0001323349,0.00009757349,0.00064195,0.00002460154,0.0000336126,0.0002152189,0.0003927097],"category_scores_gemma":[0.000072828,0.0001131315,0.00002374979,0.0002590283,0.0002855338,0.0001617138,0.00002434857,0.0001471498,0.0002905385],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000302832,"about_ca_system_score_gemma":0.00002172828,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003217969,"about_ca_topic_score_gemma":0.00005649931,"domain_scores_codex":[0.999108,0.0001536553,0.0002034618,0.0002458336,0.00007672541,0.0002123899],"domain_scores_gemma":[0.9996182,0.0001433943,0.00008755367,0.00004152234,0.00008469866,0.00002460778],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","study_design_scores_codex":[0.0007304802,0.0007099101,0.2610091,0.0000813723,0.0001562691,0.00009234814,0.2521317,2.500359e-7,0.007295051,0.2635885,0.04777459,0.1664304],"study_design_scores_gemma":[0.001329617,0.00003847034,0.9663383,0.000006380668,0.0000226349,0.00000719904,0.004379184,0.000004798845,0.000007732226,0.02723566,0.0005076376,0.0001224395],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9911046,0.00009652514,0.00002194055,0.001133318,0.0002121463,0.0002458202,0.00007528306,0.0001131973,0.006997176],"genre_scores_gemma":[0.9974693,0.00001336737,0.000002995336,0.0002586727,0.0002633748,0.00007375224,0.0004379271,0.00001312571,0.001467504],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7053291,"threshold_uncertainty_score":0.4937425,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06644514281562336,"score_gpt":0.3790856941943569,"score_spread":0.3126405513787336,"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."}}