{"id":"W4404993326","doi":"10.2196/57451","title":"Performance of GPT-3.5 and GPT-4 on the Korean Pharmacist Licensing Examination: Comparison Study","year":2024,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Pharmacy; Pharmacist; Medicine; Legislation; Medical education; Family medicine; Political science","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":[0.005346777,0.001432853,0.0008595707,0.001688943,0.0003481099,0.0009659956,0.001104975,0.001367509,0.002083344],"category_scores_gemma":[0.0260247,0.0003820968,0.001229229,0.0006518513,0.0003618195,0.001733655,0.00191768,0.001020581,0.001390463],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001280012,"about_ca_system_score_gemma":0.001396656,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01118493,"about_ca_topic_score_gemma":0.008264217,"domain_scores_codex":[0.9966117,0.001686881,0.0002686305,0.0006451816,0.0005547151,0.0002328995],"domain_scores_gemma":[0.9852528,0.008842578,0.0009743254,0.001027018,0.002738215,0.001164964],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01269845,0.002941966,0.4783852,0.001405836,0.0009772761,0.0008347451,0.002754572,0.0413088,0.008015708,0.0005355948,0.01116628,0.4389755],"study_design_scores_gemma":[0.0005932226,0.007456235,0.4314271,0.0002889096,0.001064861,0.001204176,0.002320838,0.5331886,0.01356934,0.001099721,0.007518439,0.0002685663],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9903225,0.000436792,0.004313384,0.0002451749,0.00008600443,0.0002177132,0.001253015,0.001212774,0.001912452],"genre_scores_gemma":[0.9873391,0.0002416754,0.007319427,0.0001315427,0.00003570646,0.0002041277,0.002935801,0.00007208662,0.001720584],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01118493,"threshold_uncertainty_score":0.0282768,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.15283647417708,"score_gpt":0.4955732877592086,"score_spread":0.3427368135821286,"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."}}