{"id":"W4392982637","doi":"10.1038/s41433-024-03037-w","title":"ChatGPT-3.5 and Bing Chat in ophthalmology: an updated evaluation of performance, readability, and informative sources","year":2024,"lang":"en","type":"article","venue":"Eye","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":false,"ca_institutions":"St. Michael's Hospital; Kensington Health; University of Toronto; University Health Network; Toronto Western Hospital; Canada Research Chairs; University of British Columbia","funders":"","keywords":"Readability; Grade level; MEDLINE; Medicine; Citation; Computer science; Medical education; Information retrieval; World Wide Web; Psychology; Mathematics education","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"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":[0.01991434,0.001109541,0.001899355,0.005837298,0.0007074279,0.002663275,0.001634575,0.001572005,0.003966239],"category_scores_gemma":[0.1236977,0.0007088874,0.001875483,0.00294956,0.000666093,0.004233038,0.003476012,0.001957354,0.002962315],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001299897,"about_ca_system_score_gemma":0.002394853,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009756887,"about_ca_topic_score_gemma":0.01253077,"domain_scores_codex":[0.9831167,0.005942666,0.002500048,0.0008457957,0.007045288,0.0005494379],"domain_scores_gemma":[0.7943652,0.1452,0.01601757,0.006317282,0.0304581,0.00764184],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0174635,0.003787491,0.4881442,0.004333325,0.002430478,0.0002851274,0.004290943,0.001747984,0.002482486,0.0003724779,0.04894978,0.4257122],"study_design_scores_gemma":[0.001316298,0.005350254,0.955905,0.001617967,0.002343957,0.001021207,0.00149394,0.006101659,0.005130927,0.0005151763,0.01882391,0.0003798146],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9367846,0.009853384,0.004902266,0.002142204,0.0006673169,0.001690932,0.0243383,0.004822291,0.01479878],"genre_scores_gemma":[0.9230842,0.006913152,0.01878119,0.001721859,0.0004539956,0.003121888,0.03504872,0.00146018,0.009414797],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01991434,"threshold_uncertainty_score":0.1053184,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1714382149234133,"score_gpt":0.4659599381342023,"score_spread":0.2945217232107891,"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."}}