{"id":"W4406827449","doi":"10.1016/j.jcjo.2025.01.001","title":"Assessing the performance of Microsoft Copilot, GPT-4 and Google Gemini in ophthalmology","year":2025,"lang":"en","type":"article","venue":"Canadian Journal of Ophthalmology","topic":"Retinal Imaging and Analysis","field":"Medicine","cited_by":20,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto; Centre Hospitalier de l’Université de Montréal; Université de Montréal","funders":"Université de Montréal; University of Toronto","keywords":"Computer science; GCM transcription factors; Operations research; Aeronautics; Mathematics; Engineering; Biology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.003018915,0.0005373368,0.0004786298,0.001981877,0.0002760487,0.0019851,0.0006420133,0.001085164,0.001588639],"category_scores_gemma":[0.03062783,0.0002172008,0.0005445529,0.00153841,0.0003962757,0.001623122,0.0008680389,0.0006805813,0.001186552],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009005706,"about_ca_system_score_gemma":0.0008420163,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01898023,"about_ca_topic_score_gemma":0.01891388,"domain_scores_codex":[0.9974123,0.0007659806,0.0003031181,0.0003054658,0.0009781647,0.0002350144],"domain_scores_gemma":[0.9832025,0.01013985,0.001630075,0.0005156306,0.003068999,0.001442957],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.02152676,0.001398741,0.5912647,0.00131366,0.0006786611,0.0008187204,0.004208154,0.01017438,0.007727394,0.0009152507,0.01856398,0.3414096],"study_design_scores_gemma":[0.0003203944,0.005278284,0.9256834,0.0003414875,0.0006434637,0.001759911,0.00408672,0.03947605,0.01192246,0.0008067991,0.009469369,0.0002115829],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9917541,0.001453225,0.0008182072,0.000457581,0.00006240172,0.00004870119,0.0009137493,0.0005344682,0.003957467],"genre_scores_gemma":[0.9942754,0.0006030935,0.002727388,0.0001243469,0.00004343944,0.00001764088,0.0009133859,0.0001072492,0.00118815],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01898023,"threshold_uncertainty_score":0.03773952,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02403652850803762,"score_gpt":0.3370729267759948,"score_spread":0.3130363982679571,"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."}}