{"id":"W4401322393","doi":"10.1016/j.aucc.2024.07.036","title":"Evaluating the appropriateness, consistency, and readability of ChatGPT in critical care recommendations","year":2024,"lang":"en","type":"article","venue":"Australian Critical Care","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Toronto; Western University","funders":"","keywords":"Medicine; Readability; Consistency (knowledge bases); Appropriateness criteria; Medical physics; Linguistics; Radiology; Artificial intelligence","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000475796,0.0001064147,0.0002011836,0.00007075465,0.0001277703,0.00004670287,0.00005959009,0.0001321566,0.000226033],"category_scores_gemma":[0.00357929,0.00007975186,0.00006011352,0.0002540135,0.0005318581,0.0001006788,0.00002880397,0.0003721953,0.00001325239],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001316683,"about_ca_system_score_gemma":0.0003864843,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001176994,"about_ca_topic_score_gemma":0.0002868773,"domain_scores_codex":[0.9985668,0.0001729647,0.0004645155,0.0003170834,0.0002172698,0.0002613864],"domain_scores_gemma":[0.9976983,0.001326359,0.00001379508,0.0002650568,0.000565112,0.0001313536],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001960906,0.0002635183,0.1117744,0.02074618,0.00003144049,0.00007898182,0.07805691,0.000008085307,0.0019928,0.0905195,0.0008787062,0.6954534],"study_design_scores_gemma":[0.0001448978,0.002798029,0.06249969,0.005012054,0.0008144829,0.000235925,0.8432093,0.002089111,0.0332035,0.04524441,0.004065532,0.000683064],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8938346,0.003600653,0.00003976755,0.100248,0.000802555,0.0005028427,0.00006162266,0.00004607014,0.0008638884],"genre_scores_gemma":[0.9983256,0.00002839783,0.001077952,0.0002232476,0.0001727276,0.00006065594,0.00004825498,0.00001204229,0.00005107141],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7651524,"threshold_uncertainty_score":0.4285003,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4545252341154738,"score_gpt":0.5849867038464664,"score_spread":0.1304614697309927,"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."}}