{"id":"W7101382243","doi":"10.1093/eurpub/ckaf161.316","title":"Large language models to reduce antimicrobial resistance: ChatGPT, Claude and Gemini comparison","year":2025,"lang":"en","type":"article","venue":"European Journal of Public Health","topic":"Antibiotic Use and Resistance","field":"Immunology and Microbiology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Readability; Quality (philosophy); Public health; Diversity (politics); Lexical diversity; Health informatics; Resistance (ecology)","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":[],"consensus_categories":[],"category_scores_codex":[0.0101726,0.001201585,0.001122241,0.001701826,0.0004171316,0.001549017,0.0008212395,0.001083622,0.004390312],"category_scores_gemma":[0.03733113,0.0002399681,0.001251963,0.0007015303,0.0004711167,0.001917954,0.001313636,0.001516648,0.0007119902],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001079499,"about_ca_system_score_gemma":0.0009637461,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002880086,"about_ca_topic_score_gemma":0.002127191,"domain_scores_codex":[0.996242,0.002886193,0.0002470133,0.0002518367,0.0002881163,0.00008480853],"domain_scores_gemma":[0.9577003,0.0387,0.0006888316,0.0009684607,0.001190544,0.0007517925],"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":[0.03246121,0.008577084,0.04083355,0.009044665,0.00421147,0.0006153343,0.003387172,0.1967204,0.006799626,0.00836669,0.02928291,0.6596999],"study_design_scores_gemma":[0.003627135,0.02150781,0.04250588,0.0006751214,0.00300186,0.0004231507,0.001605462,0.8974054,0.005007118,0.01126738,0.01276079,0.0002128357],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9362977,0.009921808,0.03380238,0.001685653,0.0005833472,0.001124863,0.00225729,0.002379722,0.01194729],"genre_scores_gemma":[0.9662902,0.001536587,0.02538148,0.0002661563,0.0001560045,0.0006755101,0.0031001,0.0001798943,0.002414031],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0101726,"threshold_uncertainty_score":0.0537985,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03321511516223329,"score_gpt":0.3082562961451052,"score_spread":0.2750411809828719,"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."}}