{"id":"W4401171954","doi":"10.1371/journal.pone.0307383","title":"Evaluation of ChatGPT as a diagnostic tool for medical learners and clinicians","year":2024,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":33,"is_retracted":false,"has_abstract":true,"ca_institutions":"London Health Sciences Centre; Western University","funders":"","keywords":"Correctness; Cognitive load; Medical diagnosis; Qualitative research; Cognition; Diagnostic accuracy; Computer science; Qualitative analysis; Medical education; Quality (philosophy); Resource (disambiguation); Medicine; Medical physics; Data science; Pathology; Radiology; Psychiatry","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":[0.05494922,0.0007499694,0.001275491,0.0024832,0.001551148,0.003736748,0.00194366,0.001360553,0.00402006],"category_scores_gemma":[0.2552618,0.0005130933,0.001068091,0.001221885,0.001818191,0.003448777,0.006680468,0.002022745,0.00125617],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004357733,"about_ca_system_score_gemma":0.00536721,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001594413,"about_ca_topic_score_gemma":0.002421149,"domain_scores_codex":[0.9552494,0.03085629,0.003611184,0.001822972,0.006909078,0.001551019],"domain_scores_gemma":[0.6866738,0.2359762,0.01798978,0.007277843,0.03867822,0.01340427],"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":[0.004321945,0.004463493,0.3836943,0.005217678,0.0002397109,0.002849549,0.1763739,0.001695578,0.003533044,0.001171135,0.01745796,0.3989817],"study_design_scores_gemma":[0.002079078,0.02616998,0.5902045,0.008176249,0.0008862996,0.008112505,0.2526392,0.03127948,0.01326436,0.004591839,0.06164343,0.0009530958],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9819125,0.0003651849,0.005781006,0.001723942,0.0002005949,0.004190146,0.0006329477,0.0003812625,0.004812392],"genre_scores_gemma":[0.9669845,0.0004941928,0.02200899,0.0008888797,0.000140434,0.007602654,0.0005463261,0.000076631,0.001257457],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05494922,"threshold_uncertainty_score":0.2906027,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3661206008068268,"score_gpt":0.4925521632933395,"score_spread":0.1264315624865126,"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."}}