{"id":"W4386407038","doi":"10.2196/46482","title":"Artificial Intelligence in Medical Education: Comparative Analysis of ChatGPT, Bing, and Medical Students in Germany","year":2023,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":109,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medical education; German; Psychology; Medicine; Geography","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002644392,0.0003138432,0.0004112418,0.002798911,0.0004481192,0.001439646,0.0003790679,0.0005629231,0.001349243],"category_scores_gemma":[0.009339203,0.0001621237,0.0003518069,0.0018075,0.0007264754,0.0007742606,0.001391159,0.000459925,0.0003241285],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001432104,"about_ca_system_score_gemma":0.001335639,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009346765,"about_ca_topic_score_gemma":0.01127674,"domain_scores_codex":[0.9985697,0.0004784819,0.000144319,0.0001665457,0.0003816222,0.0002592732],"domain_scores_gemma":[0.9953104,0.001888072,0.0009041321,0.0001811533,0.0004721814,0.001244121],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002059307,0.0002803752,0.9721071,0.000086474,0.00005828457,0.0003007246,0.004982734,0.0002911251,0.0003932083,0.0002184335,0.0004936737,0.02058199],"study_design_scores_gemma":[0.000006004132,0.0002275411,0.994467,0.00002657581,0.00002020139,0.0001493007,0.003342334,0.0007620153,0.0002067321,0.00006158835,0.000722627,0.000007892139],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9992728,0.0001443006,0.0000254471,0.0000917647,0.000004202743,0.000006499957,0.00005931033,0.000003074379,0.0003926171],"genre_scores_gemma":[0.9994005,0.0001363686,0.00005568843,0.00003247949,0.000004547379,0.000007020814,0.00009484384,0.000001563311,0.0002670392],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.009346765,"threshold_uncertainty_score":0.01858467,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1132779861705955,"score_gpt":0.5313391288631245,"score_spread":0.418061142692529,"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."}}