{"id":"W4360796975","doi":"10.1101/2023.03.16.23286914","title":"Learning to fake it: limited responses and fabricated references provided by ChatGPT for medical questions","year":2023,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":32,"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":"Quality (philosophy); Quartile; Psychology; Medical education; Interpretation (philosophy); Minor (academic); Computer science; Medicine; Humanities","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":[0.08872502,0.001069775,0.001073067,0.003237405,0.001935592,0.003383962,0.001449825,0.003276301,0.00541418],"category_scores_gemma":[0.4909689,0.0006107163,0.0008202231,0.001697347,0.00272541,0.002830482,0.00450457,0.002014702,0.001847927],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002305782,"about_ca_system_score_gemma":0.001743124,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006528766,"about_ca_topic_score_gemma":0.0007628826,"domain_scores_codex":[0.8239495,0.1370989,0.01232273,0.005572463,0.01868291,0.002373485],"domain_scores_gemma":[0.3108258,0.5675534,0.05888678,0.02005798,0.03968232,0.002993763],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.005115209,0.0006669479,0.160957,0.01050279,0.0004831969,0.008336156,0.4166796,0.001258434,0.02516796,0.004302837,0.02459333,0.3419365],"study_design_scores_gemma":[0.0008936824,0.01020423,0.3544753,0.0221032,0.001383961,0.02312581,0.3305937,0.01951828,0.06365894,0.01754968,0.1550639,0.001429367],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9491595,0.002710796,0.02824863,0.008086264,0.0008969845,0.001072734,0.0006392553,0.0006897163,0.008496176],"genre_scores_gemma":[0.9670473,0.001253888,0.02142781,0.004486026,0.0004855352,0.002055282,0.0003636219,0.0003006292,0.002579931],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9967237,"threshold_uncertainty_score":0.4692283,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2205691592056989,"score_gpt":0.4621102855995091,"score_spread":0.2415411263938102,"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."}}