{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001035785,0.00005382363,0.0001109291,0.0001654082,0.0000352203,0.00001343092,0.00001820134,0.00006071784,0.00008414483],"category_scores_gemma":[0.00007595542,0.00004479814,0.00000795607,0.0001808343,0.00008388273,0.0002814818,0.00001520531,0.0001033607,0.000004974489],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004404676,"about_ca_system_score_gemma":0.0001224215,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003085528,"about_ca_topic_score_gemma":0.00003222456,"domain_scores_codex":[0.9993712,0.00005349127,0.0002289088,0.0001159946,0.0001368511,0.00009359484],"domain_scores_gemma":[0.99968,0.00004419392,0.00003191491,0.00008531207,0.0001137854,0.0000447997],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00007747907,0.00005224474,0.6132139,0.0004270278,0.00001352809,0.000002357291,0.0447236,0.00003364788,0.0004889829,0.0002024753,0.00002058157,0.3407442],"study_design_scores_gemma":[0.0001083392,0.0009124054,0.7608134,0.000588192,0.00008347593,0.0001056671,0.01852774,0.1936644,0.0214213,0.003150632,0.0004808555,0.0001435495],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9972272,0.0007573019,0.000003821657,0.0006252119,0.0001286485,0.0002789682,0.000001387527,0.00001526004,0.000962196],"genre_scores_gemma":[0.9996576,0.000105771,0.00006318132,0.00003581197,0.00004460941,0.00001532841,0.00003346542,0.00000440384,0.00003982573],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3406006,"threshold_uncertainty_score":0.1826815,"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."}}