{"id":"W4394760737","doi":"10.1001/jamaophthalmol.2024.0017","title":"Accuracy of an Artificial Intelligence Chatbot’s Interpretation of Clinical Ophthalmic Images","year":2024,"lang":"en","type":"article","venue":"JAMA Ophthalmology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":83,"is_retracted":false,"has_abstract":true,"ca_institutions":"Sunnybrook Health Science Centre; St. Michael's Hospital; McMaster University; University of Toronto","funders":"","keywords":"Chatbot; Medicine; Artificial intelligence; Medical physics; Ophthalmology; Optometry; Computer science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.01906547,0.0006808826,0.0005783197,0.001466581,0.0004519325,0.002037386,0.001257454,0.001580118,0.003437246],"category_scores_gemma":[0.1115398,0.0003918566,0.0007647017,0.0005690291,0.0009304628,0.001753257,0.001773872,0.000890348,0.002057703],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001100415,"about_ca_system_score_gemma":0.0008850624,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003747325,"about_ca_topic_score_gemma":0.003371307,"domain_scores_codex":[0.9852299,0.007539301,0.001420284,0.002345844,0.002767839,0.0006968093],"domain_scores_gemma":[0.8864217,0.08035661,0.01525325,0.005417414,0.01017474,0.002376397],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002795283,0.0007108583,0.8957989,0.0005493079,0.0003109784,0.0004383829,0.004008717,0.004739272,0.005184399,0.0002513497,0.00568557,0.07952692],"study_design_scores_gemma":[0.0001103936,0.002806208,0.8931817,0.0005552383,0.0002633647,0.001799176,0.003607718,0.0788009,0.0124779,0.000661298,0.005574059,0.0001620078],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9903582,0.000277511,0.003903634,0.0005148565,0.00007966406,0.0001411059,0.0008397684,0.0005727933,0.003312466],"genre_scores_gemma":[0.9945497,0.00006610435,0.003421586,0.0001968052,0.00003060857,0.00005471006,0.0008210325,0.0000467872,0.0008126188],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01906547,"threshold_uncertainty_score":0.1008291,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2574401786806904,"score_gpt":0.5350129649838523,"score_spread":0.277572786303162,"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."}}