{"id":"W4411354886","doi":"10.1177/11206721251350809","title":"Artificial intelligence versus ophthalmology experts: Comparative analysis of responses to blepharitis patient queries","year":2025,"lang":"en","type":"article","venue":"European Journal of Ophthalmology","topic":"Retinal and Optic Conditions","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Blepharitis; Correctness; Likert scale; Context (archaeology); Medicine; Test (biology); Computer science; Psychology; Dermatology","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":[0.01488548,0.0002918413,0.0005162085,0.001469101,0.0002666151,0.0009895656,0.000431525,0.0007811018,0.001461721],"category_scores_gemma":[0.1123747,0.0001185855,0.0004526502,0.001011953,0.000530614,0.0009330827,0.00134473,0.0004907914,0.0004046101],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004747595,"about_ca_system_score_gemma":0.0004598849,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00082221,"about_ca_topic_score_gemma":0.0005361639,"domain_scores_codex":[0.9759651,0.01717305,0.001968518,0.001121475,0.003030758,0.0007411407],"domain_scores_gemma":[0.8479165,0.1337701,0.008247889,0.001752904,0.006719642,0.00159294],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004658514,0.0005281117,0.8733079,0.0006041388,0.0003064831,0.0005054353,0.02583981,0.001698285,0.002494735,0.0003104887,0.001758891,0.08798718],"study_design_scores_gemma":[0.000209941,0.003940075,0.94104,0.0003003801,0.0002159347,0.001347499,0.0310391,0.01405095,0.002660593,0.0006632693,0.004445049,0.0000873311],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.997447,0.0001502036,0.0006537765,0.0002053482,0.00001786721,0.00006848732,0.0001459059,0.00001972354,0.001291839],"genre_scores_gemma":[0.9986266,0.00009440984,0.0006276926,0.0001692427,0.00001486323,0.00007217649,0.0002230968,0.000007200032,0.0001647471],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01488548,"threshold_uncertainty_score":0.07872283,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1033451015112932,"score_gpt":0.3939042094004439,"score_spread":0.2905591078891507,"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."}}