{"id":"W4405650006","doi":"10.1016/j.jfo.2024.104391","title":"Performance of ChatGPT in French language analysis of multimodal retinal cases","year":2024,"lang":"fr","type":"article","venue":"Journal Français d Ophtalmologie","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Sunnybrook Health Science Centre; Hôpital Maisonneuve-Rosemont; St. Michael's Hospital; Université de Montréal; McGill University; University of Toronto","funders":"","keywords":"Retinal; Computer science; Linguistics; Natural language processing; Artificial intelligence; Ophthalmology; Medicine; Philosophy","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001488736,0.0002113663,0.000763527,0.001344716,0.00007298894,0.00002823015,0.0002059674,0.0003275444,0.002668599],"category_scores_gemma":[0.001147536,0.0001841268,0.000400135,0.002329958,0.0003315337,0.0002609925,0.00004146139,0.0009542992,0.00007104838],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002765153,"about_ca_system_score_gemma":0.0006999093,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01681973,"about_ca_topic_score_gemma":0.0003918658,"domain_scores_codex":[0.9972115,0.0001887694,0.001391089,0.0002762536,0.0004197262,0.0005126719],"domain_scores_gemma":[0.9978852,0.001046015,0.0002846435,0.0002847522,0.0003132863,0.0001861021],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002381386,0.0005277983,0.6745986,0.00159622,0.000780465,0.00287327,0.02428841,0.002660502,0.003413565,0.00004836529,0.000581049,0.2883936],"study_design_scores_gemma":[0.0001611877,0.003145719,0.7668289,0.005832067,0.002256735,0.002525011,0.01219176,0.1828358,0.02282782,0.0001261871,0.0009015942,0.0003671516],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8825665,0.1105356,0.00007016912,0.001687752,0.004334036,0.0001617455,0.00004603227,0.00001309526,0.0005850096],"genre_scores_gemma":[0.9891054,0.008193002,0.0006384298,0.00005228127,0.001060147,0.000006589114,0.00001954842,0.00001903077,0.0009055598],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2880265,"threshold_uncertainty_score":0.9982431,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09831981676318398,"score_gpt":0.4274383736009609,"score_spread":0.3291185568377769,"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."}}