{"id":"W4399469626","doi":"10.2196/51383","title":"Optimizing ChatGPT’s Interpretation and Reporting of Delirium Assessment Outcomes: Exploratory Study","year":2024,"lang":"en","type":"article","venue":"JMIR Formative Research","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Trillium Health Centre; University of Toronto","funders":"National Institute on Aging; National Institutes of Health; Emory University","keywords":"Interpretation (philosophy); Delirium; Exploratory research; Psychology; Medicine; Computer science; Psychiatry; Sociology; Social science","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":[0.03282309,0.001058848,0.0008459118,0.001189644,0.0004474555,0.001389729,0.001905922,0.0008237467,0.001425083],"category_scores_gemma":[0.1536956,0.0005996061,0.001002934,0.0006092097,0.0008636037,0.001405556,0.002430284,0.00144233,0.0006830348],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001713886,"about_ca_system_score_gemma":0.002738759,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002375738,"about_ca_topic_score_gemma":0.002625037,"domain_scores_codex":[0.9764066,0.01732147,0.002052349,0.001761509,0.001777805,0.0006803015],"domain_scores_gemma":[0.8120629,0.1509964,0.009013816,0.01028607,0.0151687,0.002472019],"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":[0.00788547,0.01623639,0.328442,0.002425611,0.0005807312,0.001238195,0.04542188,0.03769336,0.01536956,0.001011559,0.00420075,0.5394945],"study_design_scores_gemma":[0.002501474,0.05645055,0.4788713,0.001265986,0.001749602,0.00205367,0.02149079,0.3323636,0.08225527,0.004595575,0.01549796,0.0009042058],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9705626,0.00007238752,0.02569697,0.0001014008,0.00001768222,0.001982868,0.0002965328,0.0006179999,0.0006515047],"genre_scores_gemma":[0.9291559,0.00008860225,0.06623471,0.0001198779,0.00001836749,0.003274586,0.0005586403,0.0001137451,0.0004355195],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03282309,"threshold_uncertainty_score":0.1735871,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4140364371140245,"score_gpt":0.6073678527791601,"score_spread":0.1933314156651355,"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."}}