{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.001646729,0.0001826572,0.0004101907,0.0003028235,0.0003844647,0.0001459237,0.0001232557,0.0002045101,0.00004223544],"category_scores_gemma":[0.01592723,0.000163931,0.00005983306,0.0009146145,0.000098755,0.0002585562,0.00006654083,0.000401525,0.0002198078],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001346174,"about_ca_system_score_gemma":0.001025741,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002343173,"about_ca_topic_score_gemma":0.00003352306,"domain_scores_codex":[0.9974673,0.00002528889,0.0009156878,0.0005468862,0.0004323839,0.0006124665],"domain_scores_gemma":[0.9972036,0.001110555,0.0002224769,0.000103344,0.0004623885,0.0008976042],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001356058,0.0003249446,0.05569852,0.001226178,0.00003786958,0.000003476247,0.0112037,2.716901e-7,0.00009308097,0.0004488012,0.2189863,0.7106208],"study_design_scores_gemma":[0.001001439,0.02194541,0.04057065,0.00439577,0.00008022798,0.000155364,0.07679661,0.005962503,0.001067409,0.006616178,0.8403379,0.001070512],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8319949,0.000167481,0.000130457,0.1646352,0.000275052,0.001465798,0.00004984624,0.0004745385,0.0008067113],"genre_scores_gemma":[0.9828056,0.0007672799,0.0001871028,0.005989051,0.0002894725,0.0003251446,0.0002213871,0.00003661316,0.00937837],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7095503,"threshold_uncertainty_score":0.992362,"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."}}