{"id":"W4390057118","doi":"10.1097/ana.0000000000000949","title":"Utilizing Artificial Intelligence and Chat Generative Pretrained Transformer to Answer Questions About Clinical Scenarios in Neuroanesthesiology","year":2023,"lang":"en","type":"article","venue":"Journal of Neurosurgical Anesthesiology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Medicine; Transformer; Generative grammar; Artificial intelligence; Electrical engineering; Engineering; Computer science","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":[0.001315672,0.0001939401,0.0007237734,0.0005599001,0.0001001529,0.00002011057,0.0001427146,0.0003069862,0.0000559376],"category_scores_gemma":[0.001552371,0.0001486504,0.0001675731,0.0006995521,0.0002989441,0.0001033862,0.00001944538,0.0007809196,0.000057637],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003520857,"about_ca_system_score_gemma":0.0003950936,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003328077,"about_ca_topic_score_gemma":0.00004288992,"domain_scores_codex":[0.9968003,0.0006016355,0.001600412,0.0003712789,0.000193572,0.0004328235],"domain_scores_gemma":[0.9977655,0.001135034,0.0002241064,0.0001681469,0.0002765207,0.0004306727],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.004109337,0.001212611,0.2633895,0.0001001015,0.00003515538,0.003602114,0.00360061,0.0009089442,0.007975429,0.005014309,0.0005571719,0.7094947],"study_design_scores_gemma":[0.0002171213,0.01270307,0.9443123,0.0004120553,0.00008961585,0.007169111,0.002337956,0.00360138,0.00296328,0.01006417,0.01572541,0.0004045668],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9809785,0.0002149016,0.0002110569,0.01734746,0.000773778,0.0003889783,8.98228e-7,0.00003054565,0.00005388532],"genre_scores_gemma":[0.9920251,0.001489064,0.0005417881,0.005184813,0.0006523676,0.00001859048,0.000003515216,0.00002145093,0.00006336047],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7090902,"threshold_uncertainty_score":0.6061789,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2525050550119966,"score_gpt":0.4513680055123303,"score_spread":0.1988629505003336,"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."}}