{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003986083,0.000197464,0.0003991063,0.001359188,0.0003036523,0.00007227002,0.0001827463,0.0001058272,0.00005842869],"category_scores_gemma":[0.003104231,0.0001717657,0.00005792236,0.001790483,0.000350864,0.0004833298,0.0003843418,0.0009851272,0.00001375906],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001804558,"about_ca_system_score_gemma":0.001794655,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009876527,"about_ca_topic_score_gemma":0.001294501,"domain_scores_codex":[0.9967352,0.000670508,0.0004626763,0.0005467384,0.0009867109,0.0005981971],"domain_scores_gemma":[0.9977152,0.0008824036,0.0001052076,0.0005643893,0.0004837795,0.0002489919],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01497305,0.01773518,0.894905,0.002738061,0.0001502738,0.00006136252,0.05679597,0.0001062542,0.003108557,0.007309254,0.0007163999,0.001400641],"study_design_scores_gemma":[0.009840342,0.007814251,0.9746387,0.0002748909,0.000009386324,0.000006364557,0.003546101,0.001163293,0.001273614,0.001235652,0.00007169595,0.0001257278],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9856229,0.0003319368,0.0001460753,0.0006683689,0.0002368145,0.003274112,0.00006353317,0.00005557057,0.009600653],"genre_scores_gemma":[0.9990178,0.0000327072,0.00001182945,0.0000534393,0.00008943032,0.000251619,0.00001686393,0.00001551962,0.0005107801],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07973369,"threshold_uncertainty_score":0.70044,"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."}}