{"id":"W4416373856","doi":"10.2196/72839","title":"ChatGPT Versus DeepSeek for Breast Cancer Information Retrieval: Quantitative Comparative Study","year":2025,"lang":"en","type":"article","venue":"JMIR Cancer","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Breast cancer; Quality assurance; Quality (philosophy); MEDLINE; Cancer","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":[0.0001630391,0.0001252746,0.0002564584,0.0001608942,0.0001681633,0.0000308781,0.00006685946,0.00007420967,0.0002190312],"category_scores_gemma":[0.00007515268,0.0001110759,0.00005923989,0.0004895417,0.00005485427,0.0003093408,0.00001531857,0.0001472384,0.0000365645],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004983206,"about_ca_system_score_gemma":0.0007523713,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002341005,"about_ca_topic_score_gemma":0.00187211,"domain_scores_codex":[0.9990171,0.0000329923,0.000372533,0.0001777129,0.0001792527,0.0002204154],"domain_scores_gemma":[0.998598,0.0002206522,0.0001309608,0.0001657384,0.0008091846,0.00007545722],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.06447761,0.002026183,0.1846915,0.00169043,0.001511456,0.000003187205,0.2345925,0.0007692548,0.000711006,0.008737179,0.09965102,0.4011387],"study_design_scores_gemma":[0.004553425,0.004456011,0.5879272,0.00132736,0.0008267963,0.000003933655,0.2744078,0.01128402,0.01333123,0.001317574,0.09976839,0.000796207],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9817013,0.0006226542,0.0009635105,0.008580361,0.002836835,0.003802649,0.0001648612,0.00007081941,0.001257069],"genre_scores_gemma":[0.9964651,0.0001250291,0.0001386334,0.0008528912,0.0002463579,0.001569521,0.00003520812,0.000007636945,0.0005596228],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4032357,"threshold_uncertainty_score":0.4529545,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2513621827024224,"score_gpt":0.5533407666584808,"score_spread":0.3019785839560584,"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."}}