{"id":"W4402548672","doi":"10.1093/pch/pxae062","title":"Will ChatGPT-4 improve the quality of medical abstracts?","year":2024,"lang":"en","type":"article","venue":"Paediatrics & Child Health","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Montreal Children's Hospital; Université de Montréal; Centre Hospitalier Universitaire Sainte-Justine","funders":"","keywords":"Quality (philosophy); Computer science; Medicine; Philosophy","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1702937,0.001250457,0.001963074,0.003282371,0.001562314,0.004798299,0.003121069,0.002441699,0.01450273],"category_scores_gemma":[0.4750825,0.0009863793,0.003722658,0.002252268,0.002096011,0.004243255,0.004934688,0.002418129,0.003416599],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002786943,"about_ca_system_score_gemma":0.007547873,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006934822,"about_ca_topic_score_gemma":0.001308799,"domain_scores_codex":[0.8333093,0.1192091,0.02012338,0.004646946,0.02086513,0.001846211],"domain_scores_gemma":[0.3007323,0.5187308,0.08690619,0.03164023,0.04932184,0.01266858],"domain_codex":null,"domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.03011993,0.003721478,0.04732878,0.05173189,0.002347137,0.000665708,0.01683852,0.001208741,0.0145021,0.001255976,0.04620613,0.7840736],"study_design_scores_gemma":[0.04535734,0.1095001,0.4699264,0.04154607,0.0120195,0.009402588,0.01699266,0.01320233,0.04465398,0.01975289,0.2148756,0.002770626],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8096312,0.03327239,0.04951015,0.0395956,0.006395596,0.02377541,0.00414505,0.008195044,0.0254795],"genre_scores_gemma":[0.7669569,0.0103619,0.1692587,0.009169087,0.00392502,0.03139933,0.002048086,0.001042185,0.005838758],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8297063,"threshold_uncertainty_score":0.9006099,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0980615993173963,"score_gpt":0.455525594907167,"score_spread":0.3574639955897707,"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."}}