{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003299632,0.00008477706,0.0002588751,0.0008222273,0.0003329261,0.00002367315,0.0001100097,0.00003662118,0.00002587803],"category_scores_gemma":[0.000620513,0.00006023099,0.00004971898,0.002841707,0.0001648743,0.0005267161,0.00008619952,0.000613727,0.0001165952],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001893623,"about_ca_system_score_gemma":0.0003602097,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001415372,"about_ca_topic_score_gemma":0.0001757971,"domain_scores_codex":[0.9975535,0.000600255,0.0005586546,0.0001737424,0.0007204979,0.0003933886],"domain_scores_gemma":[0.996849,0.001975355,0.0001113951,0.0002372955,0.0007546252,0.00007234592],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0005951967,0.001371617,0.2387381,0.0002839026,0.0001008391,0.00003028161,0.7192572,0.0004045775,0.0008102132,0.0003630575,0.001839548,0.03620546],"study_design_scores_gemma":[0.00006009399,0.0003947203,0.5368995,0.0001519647,0.000005043009,0.000002845597,0.453371,0.004096003,0.004436035,0.0004723057,0.00006243811,0.00004804786],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9963086,0.00002941535,0.00004251565,0.001305097,0.0003160667,0.00143321,0.000004630391,0.00003525871,0.0005252294],"genre_scores_gemma":[0.9985021,0.0001608135,0.000004550044,0.000009775193,0.0001668403,0.001033237,0.00003188411,0.000009977769,0.00008076786],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2981614,"threshold_uncertainty_score":0.2666372,"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."}}