{"id":"W4389315592","doi":"10.2196/51798","title":"Exploring the Potential of ChatGPT-4 in Predicting Refractive Surgery Categorizations: Comparative Study","year":2023,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Refractive surgery; Receiver operating characteristic; Categorization; Artificial intelligence; Random forest; Computer science; Machine learning; Medicine; Medical physics; Optometry; Ophthalmology; Cornea","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":[],"consensus_categories":[],"category_scores_codex":[0.009579004,0.00113694,0.0008843843,0.003017754,0.0003779487,0.00149512,0.00114411,0.001229672,0.002102569],"category_scores_gemma":[0.0331536,0.0003070462,0.001342438,0.001024278,0.0005671511,0.001722138,0.001732663,0.00111573,0.000852866],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001114737,"about_ca_system_score_gemma":0.00137461,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00656244,"about_ca_topic_score_gemma":0.004472367,"domain_scores_codex":[0.9963514,0.0020506,0.000302824,0.000576945,0.0004998593,0.0002184628],"domain_scores_gemma":[0.9543628,0.0372108,0.001525513,0.001847493,0.003709334,0.001344104],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006291836,0.001505834,0.755428,0.0005269031,0.00124967,0.0004128922,0.00132461,0.03487064,0.001952468,0.0004993624,0.003087906,0.19285],"study_design_scores_gemma":[0.000287954,0.006791416,0.3483888,0.0001943719,0.0009667162,0.00125456,0.002048981,0.6284736,0.006011289,0.002256871,0.003123751,0.0002016375],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9863279,0.0004221856,0.009604346,0.0002000968,0.00007179526,0.0002891838,0.001143847,0.0005041708,0.001436498],"genre_scores_gemma":[0.9874339,0.0001358632,0.01001928,0.00006720432,0.00004907956,0.0001945171,0.001548163,0.00003696589,0.0005149804],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009579004,"threshold_uncertainty_score":0.05065924,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6161224423580277,"score_gpt":0.562206736622761,"score_spread":0.05391570573526672,"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."}}