{"id":"W4406229395","doi":"10.2196/56126","title":"Proficiency, Clarity, and Objectivity of Large Language Models Versus Specialists’ Knowledge on COVID-19's Impacts in Pregnancy: Cross-Sectional Pilot Study","year":2025,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"COVID-19 Impact on Reproduction","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"","keywords":"Preprint; CLARITY; Coronavirus disease 2019 (COVID-19); Cross-sectional study; Objectivity (philosophy); 2019-20 coronavirus outbreak; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Psychology; Pregnancy; Medicine; Computer science; Infectious disease (medical specialty); Philosophy; Virology; Epistemology; Outbreak; World Wide Web","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.01063517,0.0004058095,0.000601117,0.001011242,0.0006937469,0.001723373,0.0005081098,0.0009209021,0.004373865],"category_scores_gemma":[0.03209003,0.0005982937,0.00104117,0.0004640533,0.001322009,0.001764454,0.001438348,0.00159158,0.0009119792],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007748089,"about_ca_system_score_gemma":0.0009263115,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002515876,"about_ca_topic_score_gemma":0.00257038,"domain_scores_codex":[0.9963091,0.001740215,0.0004605396,0.0005248931,0.0006117671,0.0003534217],"domain_scores_gemma":[0.972408,0.01633594,0.004623895,0.001327836,0.003324496,0.001979823],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009987577,0.00463079,0.9358906,0.0002398515,0.0001430779,0.0003124744,0.04361454,0.0002364079,0.0009928815,0.0001694034,0.0007337427,0.0120375],"study_design_scores_gemma":[0.0001163731,0.009860889,0.953748,0.0001530251,0.0001674351,0.0005210165,0.03033833,0.001266184,0.001208036,0.0002677631,0.002282034,0.00007094625],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9992331,0.00003610233,0.000101669,0.00003046683,0.000004156928,0.0000813771,0.00007291271,0.000002433949,0.0004377424],"genre_scores_gemma":[0.998717,0.0000645122,0.0002893275,0.00007429608,0.000008829019,0.0002238886,0.0001370926,0.000005799042,0.0004793794],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01063517,"threshold_uncertainty_score":0.05624479,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1941517332599753,"score_gpt":0.5469177866288609,"score_spread":0.3527660533688856,"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."}}