{"id":"W4415027259","doi":"10.2196/72918","title":"Taxonomy Portraits: Deciphering the Hierarchical Relationships of Medical Large Language Models","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Taxonomy (biology); Transparency (behavior); Health care; Medical care; Focus (optics); Public health","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"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":[0.003172024,0.0006540792,0.0004410112,0.006440948,0.001461908,0.003652153,0.001019014,0.0008372164,0.003209437],"category_scores_gemma":[0.02275897,0.0003758787,0.0009816586,0.003354014,0.001322944,0.003692397,0.002476186,0.001778763,0.001139384],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001929777,"about_ca_system_score_gemma":0.001937009,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01005172,"about_ca_topic_score_gemma":0.01527149,"domain_scores_codex":[0.9977336,0.0008417529,0.0001603133,0.0005390001,0.0005375881,0.0001877363],"domain_scores_gemma":[0.9843008,0.00943137,0.001828067,0.001637198,0.002210662,0.0005918447],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000655191,0.0003719647,0.2097346,0.001146972,0.0002805956,0.001069756,0.04188292,0.07695635,0.02124421,0.09998029,0.01955752,0.5271197],"study_design_scores_gemma":[0.00004090625,0.0002694665,0.1079862,0.0004165026,0.0001065711,0.000780702,0.01711751,0.6692953,0.005547068,0.1649407,0.03334584,0.0001531841],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4819561,0.001753957,0.4871047,0.002378958,0.00007447277,0.0007518858,0.005425904,0.002112285,0.01844172],"genre_scores_gemma":[0.7798395,0.0003662471,0.2103446,0.0001925005,0.0000334818,0.0004541671,0.006849406,0.000220004,0.001700253],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01005172,"threshold_uncertainty_score":0.01998645,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1300725114383744,"score_gpt":0.4198590540102389,"score_spread":0.2897865425718644,"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."}}