{"id":"W4414483645","doi":"10.3389/frai.2025.1618378","title":"Evaluation of large language model-generated medical information on idiopathic pulmonary fibrosis","year":2025,"lang":"en","type":"article","venue":"Frontiers in Artificial Intelligence","topic":"Interstitial Lung Diseases and Idiopathic Pulmonary Fibrosis","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Idiopathic pulmonary fibrosis; Medical information; Field (mathematics); Health care; MEDLINE; Information system","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":[0.002072154,0.0001695049,0.0003533956,0.0005188893,0.00006946665,0.0000193452,0.0001823593,0.0002294591,0.0001272907],"category_scores_gemma":[0.0009577956,0.0001613264,0.0001342301,0.0006559042,0.0001069692,0.0002235404,0.00005545171,0.0003026843,0.00002575193],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002949856,"about_ca_system_score_gemma":0.0007390542,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006438065,"about_ca_topic_score_gemma":0.00001784793,"domain_scores_codex":[0.9973595,0.0001861643,0.0007681538,0.0002434158,0.001164257,0.0002785005],"domain_scores_gemma":[0.9989778,0.00002555071,0.0001255272,0.0003122748,0.0004470718,0.0001117309],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01089106,0.002352876,0.002904504,0.0009673545,0.0002125486,0.00006416232,0.002663997,0.03222426,0.009229725,0.03652097,0.0044526,0.897516],"study_design_scores_gemma":[0.0001639803,0.0003626889,0.0003966424,0.001237177,0.0001514737,0.000002644389,0.001399798,0.9644278,0.02487133,0.00675342,0.00009814629,0.0001348704],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5593873,0.004060721,0.4112382,0.00207414,0.002745078,0.001727843,0.0001858918,0.0001319572,0.01844889],"genre_scores_gemma":[0.9982253,0.0001107372,0.0008024928,0.0005402932,0.00008311134,0.00005786605,0.000105878,0.000009623901,0.00006465153],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9322036,"threshold_uncertainty_score":0.65787,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02248705125604726,"score_gpt":0.3242837383973335,"score_spread":0.3017966871412863,"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."}}