{"id":"W4397049424","doi":"10.2196/58758","title":"Evaluation of ChatGPT-Generated Differential Diagnosis for Common Diseases With Atypical Presentation: Descriptive Research","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Presentation (obstetrics); Public health; Medicine; Differential diagnosis; Family medicine; Pathology; Computer science; Surgery","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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001307499,0.0001066367,0.0001986543,0.0002679269,0.0001304422,0.00004456746,0.00008743749,0.0001629522,0.001260957],"category_scores_gemma":[0.002232824,0.00008096041,0.00006369937,0.0006008505,0.0002295632,0.0001434918,0.00001551591,0.0002789831,0.00002279503],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003842558,"about_ca_system_score_gemma":0.006712787,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003265769,"about_ca_topic_score_gemma":0.0001398803,"domain_scores_codex":[0.9967641,0.0003626582,0.0004103116,0.0003218233,0.001906258,0.0002348508],"domain_scores_gemma":[0.9964579,0.0006149007,0.00005736956,0.0002159269,0.002317623,0.0003363045],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0009402885,0.002382807,0.02420088,0.001014823,0.0001207641,0.000001945079,0.007825363,0.000006918076,0.0002951235,0.003019423,0.04081403,0.9193776],"study_design_scores_gemma":[0.002157542,0.008744707,0.5047974,0.01468736,0.003913027,0.0001177851,0.05175056,0.2574298,0.07734296,0.04883483,0.02916037,0.001063629],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9800516,0.001995048,0.0006928319,0.01333325,0.001317099,0.002355839,0.00001376737,0.00005052746,0.0001899941],"genre_scores_gemma":[0.9924148,0.0001399425,0.0001475449,0.0001404135,0.001650353,0.004767193,0.0005286427,0.00002236421,0.0001886865],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.918314,"threshold_uncertainty_score":0.999652,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3304753861486333,"score_gpt":0.5699306547088675,"score_spread":0.2394552685602342,"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."}}