{"id":"W4396531571","doi":"10.36834/cmej.78879","title":"Large language models in medical education: new tools for experimentation and discovery","year":2024,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University","funders":"","keywords":"Computer science; Data science; Scientific discovery; Cognitive science; Psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.0477453,0.001468478,0.001559805,0.003307643,0.001215404,0.005473763,0.003631398,0.002739058,0.004727223],"category_scores_gemma":[0.1937321,0.001403889,0.002261824,0.003409942,0.005113506,0.0113737,0.007789329,0.007458518,0.001063484],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001633596,"about_ca_system_score_gemma":0.002695338,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005695915,"about_ca_topic_score_gemma":0.007698972,"domain_scores_codex":[0.9307996,0.06002524,0.001644718,0.001889977,0.005299789,0.0003406736],"domain_scores_gemma":[0.5736005,0.4066856,0.002860399,0.01261093,0.00328191,0.0009606494],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001274483,0.001745802,0.01130701,0.002860804,0.001730236,0.0003804037,0.003841218,0.1558781,0.004995191,0.2625728,0.04480376,0.5086102],"study_design_scores_gemma":[0.0003680327,0.0003604407,0.001822448,0.0004614089,0.0003227323,0.0001478199,0.0007147766,0.5191413,0.003991022,0.4369493,0.03556416,0.0001566163],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01514597,0.004889287,0.9567631,0.009061742,0.0003846943,0.0004488819,0.002151062,0.00335548,0.007799784],"genre_scores_gemma":[0.1744312,0.002407621,0.8143023,0.001545151,0.0003419785,0.001922678,0.002673585,0.0006173567,0.001758096],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.0477453,"threshold_uncertainty_score":0.2525043,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08632132941267898,"score_gpt":0.4632856320019739,"score_spread":0.3769643025892949,"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."}}