{"id":"W4380360840","doi":"10.1016/j.mcpdig.2023.05.004","title":"Learning to Fake It: Limited Responses and Fabricated References Provided by ChatGPT for Medical Questions","year":2023,"lang":"en","type":"article","venue":"Mayo Clinic Proceedings Digital Health","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":172,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Montreal Children's Hospital; Université de Montréal; Cegep de Saint Hyacinthe; Centre Hospitalier Universitaire Sainte-Justine","funders":"","keywords":"Fake news; Medical education; Psychology; Computer science; Internet privacy; Data science; Medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.04932514,0.0007648341,0.0007970697,0.002297446,0.00137659,0.002456484,0.0009923186,0.002656247,0.003328162],"category_scores_gemma":[0.4085624,0.0005417399,0.0006834877,0.001201437,0.00219478,0.002430395,0.003225069,0.001599432,0.001032159],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001645515,"about_ca_system_score_gemma":0.00118728,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008317216,"about_ca_topic_score_gemma":0.0008852527,"domain_scores_codex":[0.9033399,0.07312633,0.006861873,0.003329263,0.01171998,0.001622692],"domain_scores_gemma":[0.4925767,0.3959127,0.07262581,0.01472436,0.02134903,0.002811499],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.008284043,0.00154829,0.4827678,0.004691401,0.0005475046,0.006240309,0.2963217,0.0008437957,0.01273567,0.001674848,0.007718123,0.1766265],"study_design_scores_gemma":[0.0008780115,0.01783299,0.7007744,0.006782842,0.001122391,0.01474692,0.1788767,0.01063968,0.03054354,0.004265374,0.03277024,0.0007668816],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9920036,0.0006539082,0.003224491,0.001115759,0.0001038004,0.0003289426,0.0001859067,0.00009101868,0.00229257],"genre_scores_gemma":[0.9935142,0.0003393046,0.00378127,0.0008209226,0.00008118133,0.0005463025,0.0001331814,0.00004296317,0.0007406586],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9973438,"threshold_uncertainty_score":0.2608594,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1970116621005062,"score_gpt":0.4858427459637719,"score_spread":0.2888310838632656,"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."}}