{"id":"W4401232191","doi":"10.1016/j.jcjo.2024.06.001","title":"ChatGPT and retinal disease: a cross-sectional study on AI comprehension of clinical guidelines","year":2024,"lang":"en","type":"article","venue":"Canadian Journal of Ophthalmology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta; Toronto General Hospital; University of Toronto; Kensington Health; University Health Network; All Sum Research Center (Canada); Toronto Rehabilitation Institute","funders":"Novartis; Bayer","keywords":"Readability; Medicine; Guideline; Comprehension; Disease; Retinal; Medical physics; Family medicine; Optometry; Ophthalmology; Pathology; 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":[0.006870504,0.0002154283,0.0003839988,0.001126374,0.0005377511,0.001114272,0.000410297,0.0008820036,0.001907969],"category_scores_gemma":[0.02591706,0.0004154516,0.0005281292,0.001004388,0.0005931601,0.00129971,0.0008595847,0.001138225,0.0003449214],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006800756,"about_ca_system_score_gemma":0.0007214174,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004683312,"about_ca_topic_score_gemma":0.003674774,"domain_scores_codex":[0.9956566,0.002132758,0.0005819469,0.0003788242,0.0009359175,0.000313929],"domain_scores_gemma":[0.9721603,0.01184305,0.01035988,0.0009723391,0.002806696,0.001857594],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00005087689,0.0002157354,0.9965014,0.0000181218,0.00002630339,0.00004653927,0.001856856,0.00003132295,0.00007666094,0.00001259032,0.00009802411,0.001065572],"study_design_scores_gemma":[0.00001077497,0.0006107997,0.9954287,0.00001837491,0.00001848402,0.0002008423,0.003022507,0.0003431019,0.00005899547,0.00001707964,0.0002619902,0.000008284536],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9994714,0.00004486996,0.0000574167,0.00005431896,0.00000291854,0.00002878046,0.00007212863,0.00000233666,0.0002658869],"genre_scores_gemma":[0.9995036,0.00003934309,0.0001260989,0.00006298152,0.000004616732,0.00005075091,0.00009678544,0.00000226683,0.0001134255],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.006870504,"threshold_uncertainty_score":0.03633517,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4229977629097882,"score_gpt":0.5709980757854665,"score_spread":0.1480003128756782,"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."}}