{"id":"W4415854346","doi":"10.21203/rs.3.rs-7992691/v1","title":"From Black Box to Disposition: SurveyingDivergence, Novelty, and Ethics in Large Language Models","year":2025,"lang":"","type":"preprint","venue":"Research Square","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Saskatchewan Polytechnic","funders":"","keywords":"Face (sociological concept); Virtue; Black box; Predictive validity; Conceptual framework; Item response theory; Expression (computer science); Construct validity","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05412889,0.0005195854,0.001142586,0.002638744,0.002567303,0.007091595,0.001431307,0.002826954,0.004984684],"category_scores_gemma":[0.2441194,0.0006158645,0.00155433,0.003224757,0.009433717,0.02046222,0.007344676,0.004789741,0.0004644138],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002318089,"about_ca_system_score_gemma":0.002066268,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003228856,"about_ca_topic_score_gemma":0.003551367,"domain_scores_codex":[0.9614689,0.03031393,0.001572356,0.00232636,0.003182105,0.001136325],"domain_scores_gemma":[0.5030513,0.4280721,0.02454833,0.02731931,0.01251754,0.004491448],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.001280215,0.0008775236,0.7616323,0.0003760759,0.0005657863,0.0002140353,0.02049127,0.008494086,0.0008079741,0.1215516,0.002265851,0.08144325],"study_design_scores_gemma":[0.0002878411,0.0009073075,0.2185614,0.0004570601,0.0005454837,0.0006951438,0.02809975,0.1232792,0.002805362,0.6160664,0.008001549,0.0002935508],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9420876,0.0009558738,0.04612213,0.004132784,0.00003518457,0.00008812983,0.0003230398,0.00007058372,0.006184692],"genre_scores_gemma":[0.9949521,0.0001571157,0.004054799,0.0001883143,0.00003667792,0.00006070071,0.0002523003,0.00003401509,0.0002639675],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05412889,"threshold_uncertainty_score":0.2862644,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3843443252852912,"score_gpt":0.5696115998315502,"score_spread":0.185267274546259,"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."}}