{"id":"W4391026420","doi":"10.3390/ime3010004","title":"Examining the Efficacy of ChatGPT in Marking Short-Answer Assessments in an Undergraduate Medical Program","year":2024,"lang":"en","type":"article","venue":"International Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":27,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Rubric; Reliability (semiconductor); Medical education; Curriculum; Psychology; Computer science; Inter-rater reliability; Medicine; Mathematics education; Rating scale; Pedagogy","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":[],"consensus_categories":[],"category_scores_codex":[0.001470594,0.0001065485,0.0001636727,0.0003623939,0.00002725393,0.00004492045,0.0002678344,0.0001801326,0.0008087295],"category_scores_gemma":[0.002735786,0.00007975868,0.00004230961,0.0004805646,0.0001262754,0.0001959203,0.00003691119,0.0005815628,0.00001918651],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003140559,"about_ca_system_score_gemma":0.003114473,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008719828,"about_ca_topic_score_gemma":0.0009225163,"domain_scores_codex":[0.9974179,0.0001310402,0.0007126725,0.0002995355,0.001237741,0.0002011678],"domain_scores_gemma":[0.9987912,0.0005869756,0.00005392651,0.0001831371,0.0001654684,0.0002193253],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.00005265564,0.001247712,0.1338925,0.00007241136,0.00002017449,0.00002154223,0.001509868,0.000005424037,0.0000335864,0.002379913,0.0004090804,0.8603551],"study_design_scores_gemma":[0.0004314866,0.0009378017,0.7870061,0.009995585,0.00006829832,0.0002356078,0.00943296,0.1554241,0.0007966099,0.008532825,0.02675476,0.0003837789],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9480858,0.0003418711,0.0003286583,0.04410641,0.00425555,0.0004873958,8.573011e-7,0.00004491445,0.002348481],"genre_scores_gemma":[0.9968395,0.0002116687,0.0003717263,0.001048177,0.001041087,0.0002000874,0.0001148229,0.00001639293,0.0001565572],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8599713,"threshold_uncertainty_score":0.8855021,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1665316257906498,"score_gpt":0.5365099278259731,"score_spread":0.3699783020353232,"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."}}