{"id":"W4405763712","doi":"10.1515/tnsci-2022-0361","title":"A pilot evaluation of the diagnostic accuracy of ChatGPT-3.5 for multiple sclerosis from case reports","year":2024,"lang":"en","type":"article","venue":"Translational Neuroscience","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Wilfrid Laurier University; University of Ottawa","funders":"","keywords":"Modalities; Presentation (obstetrics); Medical physics; Medical diagnosis; Clinical Practice; Computer science; Medicine; Data science; Artificial intelligence; Machine learning; Pathology; Physical therapy; Surgery","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006373146,0.00005918407,0.00009524969,0.00005808426,0.00009458011,0.00001196552,0.00004657839,0.00002135,0.00003493908],"category_scores_gemma":[0.005340755,0.00004358259,0.00007415911,0.0003553152,0.0001339001,0.000134065,0.000006053081,0.00006831017,0.000001000313],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002055835,"about_ca_system_score_gemma":0.0005855135,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008652215,"about_ca_topic_score_gemma":0.0001816106,"domain_scores_codex":[0.9987237,0.00005072738,0.0004018156,0.0002272446,0.0005009553,0.00009549594],"domain_scores_gemma":[0.9967201,0.002610854,0.0001053886,0.000193468,0.0003285374,0.00004161861],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","study_design_scores_codex":[0.0002508915,0.0007505972,0.1056018,0.0006889011,0.00002294219,0.00008026774,0.007626779,0.009697597,0.6725762,0.001365816,0.0004840152,0.2008542],"study_design_scores_gemma":[0.0001914153,0.000463638,0.3769609,0.0009386237,0.0003499165,0.0005925018,0.0002349401,0.3562338,0.2492811,0.01413706,0.000477337,0.0001386844],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9910156,0.0005552272,0.003398883,0.002640161,0.00130735,0.0009699634,0.00006901475,0.00001436318,0.00002942483],"genre_scores_gemma":[0.999313,0.00002583873,0.0003111525,0.0001372752,0.0001020883,0.00007902503,0.00001246635,0.000006944718,0.00001223901],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4232951,"threshold_uncertainty_score":0.6393769,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4738638373127297,"score_gpt":0.4594358640871005,"score_spread":0.01442797322562922,"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."}}