{"id":"W4401538229","doi":"10.1016/j.aohep.2024.101537","title":"Evaluation of four chatbots in autoimmune liver disease: A comparative analysis","year":2024,"lang":"en","type":"article","venue":"Annals of Hepatology","topic":"Liver Diseases and Immunity","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"Chinesisch-Deutsche Zentrum für Wissenschaftsförderung; Bundesministerium für Bildung und Forschung","keywords":"Medicine; Autoimmune disease; Health professionals; Autoimmune hepatitis; Disease; Liver disease; Health care; Intensive care medicine; Pathology; Internal medicine","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":[0.02206079,0.0004666365,0.001025184,0.004341298,0.001015997,0.001531637,0.0007186151,0.0007719381,0.001457887],"category_scores_gemma":[0.07359883,0.0003177686,0.001108394,0.001985812,0.0011721,0.001881568,0.001991476,0.0006638796,0.000326735],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002380627,"about_ca_system_score_gemma":0.001795297,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001551792,"about_ca_topic_score_gemma":0.002924078,"domain_scores_codex":[0.9835414,0.01032211,0.001834775,0.0009375163,0.002544992,0.0008192336],"domain_scores_gemma":[0.8922827,0.07707633,0.008578322,0.001734045,0.01484558,0.005482921],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01368456,0.005880687,0.6152699,0.008222321,0.001169294,0.0009108299,0.05967274,0.001096557,0.004050041,0.0007446508,0.001940562,0.2873578],"study_design_scores_gemma":[0.0007073251,0.02145624,0.922779,0.001594271,0.00117541,0.0008185513,0.03769779,0.005496078,0.00254974,0.0004150337,0.005147806,0.0001627709],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9968008,0.0006485677,0.0006895598,0.00006192541,0.00002535093,0.0006246206,0.0001037443,0.00003551599,0.001009935],"genre_scores_gemma":[0.9961209,0.0004831497,0.00205819,0.00008759628,0.00002461613,0.0006430323,0.0002412136,0.00001442113,0.0003269682],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02206079,"threshold_uncertainty_score":0.11667,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2490873102605842,"score_gpt":0.4347216936920845,"score_spread":0.1856343834315002,"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."}}