{"id":"W4319454867","doi":"10.1101/2023.02.04.23285478","title":"ChatGPT for Clinical Vignette Generation, Revision, and Evaluation","year":2023,"lang":"en","type":"preprint","venue":"medRxiv","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":82,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta Hospital; University of Alberta","funders":"","keywords":"Vignette; Rewriting; Triage; Set (abstract data type); Medicine; Perspective (graphical); Literacy; Psychology; Psychiatry; Artificial intelligence; Computer science; Social psychology; Programming language","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":[],"consensus_categories":[],"category_scores_codex":[0.02114431,0.002605012,0.001253895,0.003055312,0.0006829352,0.002298418,0.003903345,0.001522853,0.0363978],"category_scores_gemma":[0.208474,0.000997983,0.001206202,0.001112731,0.0006124749,0.00273765,0.004779929,0.00181671,0.01087453],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001786991,"about_ca_system_score_gemma":0.001914096,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002080445,"about_ca_topic_score_gemma":0.002603659,"domain_scores_codex":[0.9786184,0.01478387,0.001614403,0.00215656,0.002318142,0.0005086499],"domain_scores_gemma":[0.7139776,0.240698,0.006822144,0.01416856,0.01917547,0.005158265],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.008832736,0.002721119,0.01985109,0.005523913,0.0003708705,0.001776628,0.006717375,0.01298807,0.01131733,0.003070901,0.1622752,0.7645548],"study_design_scores_gemma":[0.008387157,0.009258794,0.06745197,0.004644759,0.0009458444,0.00522513,0.005365119,0.5983186,0.0698252,0.02435094,0.2046481,0.001578377],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2402854,0.001363557,0.4078272,0.002036327,0.001233266,0.01744413,0.02284179,0.2895111,0.01745723],"genre_scores_gemma":[0.3417192,0.00052287,0.6037775,0.001241575,0.0003455669,0.01641762,0.01983769,0.008093994,0.008044015],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0363978,"threshold_uncertainty_score":0.1217627,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2538464383166623,"score_gpt":0.4922916759630908,"score_spread":0.2384452376464285,"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."}}