{"id":"W4409619030","doi":"10.2196/72522","title":"Evaluating an AI Chatbot “Prostate Cancer Info” for Providing Quality Prostate Cancer Screening Information: Cross-Sectional Study","year":2025,"lang":"en","type":"article","venue":"JMIR Cancer","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Prostate cancer; Chatbot; Medicine; Quality (philosophy); Computer science; Cancer; World Wide Web; Internal medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001478543,0.0002436431,0.0003536539,0.0002327026,0.0007236043,0.0002546596,0.0001378383,0.0001445421,0.0003277244],"category_scores_gemma":[0.0003037626,0.0002275083,0.00009950261,0.0005693194,0.00008664089,0.001301164,0.00005196411,0.000376234,0.000008170026],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009927814,"about_ca_system_score_gemma":0.002569408,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0324883,"about_ca_topic_score_gemma":0.007150307,"domain_scores_codex":[0.9972038,0.00010113,0.001170822,0.0004454206,0.0005636723,0.0005152075],"domain_scores_gemma":[0.9971318,0.0001541864,0.0004016039,0.0003284681,0.001811499,0.0001724496],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001233602,0.0001285359,0.741739,0.0006121071,0.00008711875,3.088646e-7,0.01252725,0.002558718,0.0002526326,0.00008244343,0.000665492,0.2401128],"study_design_scores_gemma":[0.001042558,0.0007293638,0.9530517,0.0006214512,0.0001049276,0.000001507005,0.008589091,0.01228924,0.003429638,0.0004619258,0.01930037,0.0003782537],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9848734,0.0006494041,0.0007257101,0.00558096,0.001980905,0.005750552,0.000169514,0.0001349945,0.0001345481],"genre_scores_gemma":[0.9760692,0.000138478,0.0005257098,0.003822492,0.001098485,0.01545304,0.0001276926,0.00002613738,0.002738791],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2397346,"threshold_uncertainty_score":0.9739544,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3580154689594601,"score_gpt":0.6265171370645302,"score_spread":0.26850166810507,"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."}}