{"id":"W4413042237","doi":"10.1097/sih.0000000000000877","title":"Exploring AI Hallucinations of ChatGPT","year":2025,"lang":"en","type":"article","venue":"Simulation in Healthcare The Journal of the Society for Simulation in Healthcare","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Alberta Children's Hospital; University of Calgary","funders":"","keywords":"Debriefing; Citation; Relevance (law); Computer science; Generative grammar; Information retrieval; Psychology; Natural language processing; Artificial intelligence; Medical education; Medicine; Library science","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":[0.003127425,0.0002016669,0.0005418382,0.0004115318,0.0004109022,0.00001744245,0.0003041823,0.0002542614,0.000007986341],"category_scores_gemma":[0.001672954,0.0001450309,0.0006314566,0.001838694,0.0001249884,0.0004083809,0.00004851148,0.0009405012,9.977518e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001075773,"about_ca_system_score_gemma":0.001434061,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002642405,"about_ca_topic_score_gemma":0.0008417539,"domain_scores_codex":[0.9960314,0.0004564267,0.002309484,0.000216458,0.0005723013,0.0004139213],"domain_scores_gemma":[0.9925205,0.003371303,0.001027439,0.0005181073,0.002442354,0.0001203724],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0005684759,0.0001393444,0.1732355,0.001495748,0.00004955903,3.091901e-7,0.01031331,0.7942873,0.00003011042,0.003032647,0.00008970968,0.01675796],"study_design_scores_gemma":[0.001245673,0.0006487106,0.2578093,0.003573564,0.0001734749,0.000008434944,0.01191559,0.684607,0.0008719358,0.03715284,0.001742601,0.0002508402],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7681448,0.002368132,0.007723677,0.2168327,0.002304472,0.00256801,0.00001849407,0.00002062544,0.00001900738],"genre_scores_gemma":[0.9946181,0.0005631281,0.0004756636,0.003853281,0.0003109552,0.00006524548,0.00001142747,0.00002580074,0.00007635929],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2264733,"threshold_uncertainty_score":0.5914189,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.291558729212726,"score_gpt":0.4918616125617453,"score_spread":0.2003028833490193,"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."}}