{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0233585,0.0005084292,0.0005134436,0.003572601,0.0005586282,0.001655499,0.001471444,0.0008993839,0.001909708],"category_scores_gemma":[0.2154302,0.000294873,0.001059251,0.001680467,0.001127282,0.001526153,0.002172467,0.001011689,0.0004157858],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002385576,"about_ca_system_score_gemma":0.001318111,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003275818,"about_ca_topic_score_gemma":0.003479591,"domain_scores_codex":[0.9874032,0.007852389,0.001192389,0.00121805,0.0020732,0.000260816],"domain_scores_gemma":[0.5996397,0.3598867,0.0114486,0.009973961,0.01756504,0.001486062],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002060208,0.0008177678,0.8800986,0.0008593364,0.0003493801,0.001344695,0.0101921,0.009325127,0.001367503,0.001542725,0.001499303,0.09054329],"study_design_scores_gemma":[0.0005466526,0.005103806,0.6482772,0.001040254,0.001033789,0.009973453,0.01780424,0.2830037,0.01755557,0.008265947,0.007150497,0.0002449686],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9902534,0.0001602167,0.006979579,0.00009930051,0.00001802053,0.0005369289,0.0008219507,0.0001203854,0.00101011],"genre_scores_gemma":[0.9890606,0.00007908462,0.009330434,0.00004860085,0.00001402893,0.0003104692,0.0009875044,0.00001853628,0.0001508048],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9766415,"threshold_uncertainty_score":0.1235331,"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."}}