{"id":"W4388206358","doi":"10.1016/j.jcjo.2023.09.011","title":"Comparison of ChatGPT to ophthalmology resident and staff consultants on an ophthalmological training tool","year":2023,"lang":"en","type":"letter","venue":"Canadian Journal of Ophthalmology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto; Western University","funders":"","keywords":"Medical education; Resource (disambiguation); Psychology; Medicine; Computer science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.006454071,0.0003383223,0.0006732185,0.002753288,0.00134534,0.002393196,0.001384956,0.001702748,0.01605049],"category_scores_gemma":[0.1273568,0.0003532025,0.0010805,0.001737971,0.0007268272,0.002544805,0.002989194,0.001742548,0.002212999],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003238908,"about_ca_system_score_gemma":0.003254127,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02106876,"about_ca_topic_score_gemma":0.02411471,"domain_scores_codex":[0.9898095,0.003781447,0.001501782,0.0005622972,0.003215971,0.001128998],"domain_scores_gemma":[0.8938072,0.06593043,0.00987209,0.002688191,0.01792603,0.00977606],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0067769,0.003340438,0.8491361,0.0007175016,0.0003379741,0.002085525,0.01594981,0.0003952113,0.0009640308,0.0005271723,0.01088081,0.1088887],"study_design_scores_gemma":[0.0002990631,0.003975485,0.9632271,0.0004172448,0.0002061483,0.001469508,0.02219141,0.001370998,0.0004946679,0.0003667526,0.005904409,0.0000771731],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9856353,0.0002464505,0.0001943588,0.001382127,0.0001822486,0.0002558191,0.0007092683,0.00006587509,0.01132855],"genre_scores_gemma":[0.9950324,0.000204927,0.0004481633,0.0007113749,0.00006894767,0.0001666462,0.0005863985,0.00004122278,0.002739971],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02106876,"threshold_uncertainty_score":0.05369425,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.38599382932281,"score_gpt":0.483549586891487,"score_spread":0.09755575756867707,"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."}}