{"id":"W4386887838","doi":"10.2196/51232","title":"Suicide Risk Assessments Through the Eyes of ChatGPT-3.5 Versus ChatGPT-4: Vignette Study","year":2023,"lang":"en","type":"article","venue":"JMIR Mental Health","topic":"Mental Health via Writing","field":"Psychology","cited_by":124,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Vignette; Mental health; Belongingness; Suicide prevention; Psychology; Poison control; Occupational safety and health; Clinical psychology; Medicine; Psychiatry; Social psychology; Medical emergency","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007375458,0.00036567,0.0004044666,0.001051216,0.001636518,0.0009728728,0.0007306937,0.001326896,0.001862162],"category_scores_gemma":[0.03675751,0.00043053,0.0005110099,0.0007632439,0.001031037,0.0009873022,0.002954276,0.001927812,0.0003390135],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002722761,"about_ca_system_score_gemma":0.0009727242,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006043911,"about_ca_topic_score_gemma":0.0117525,"domain_scores_codex":[0.9943123,0.00471367,0.0002558842,0.0001927236,0.0002755939,0.0002498015],"domain_scores_gemma":[0.9775743,0.01337518,0.001709733,0.001957289,0.002812295,0.002571299],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.004193113,0.01045427,0.3258801,0.0007036083,0.0001447506,0.01662199,0.5198005,0.005014614,0.008789799,0.003747022,0.01287137,0.09177882],"study_design_scores_gemma":[0.0004953685,0.01777195,0.5661646,0.0005380822,0.0002090744,0.01814733,0.3019785,0.02603103,0.02013046,0.003411998,0.04427842,0.0008431704],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.997956,0.0000425275,0.0006147336,0.0002148777,0.00001060577,0.0002259738,0.00007468088,0.00001144288,0.0008491981],"genre_scores_gemma":[0.9940283,0.0001104291,0.004312999,0.0001692047,0.00001966411,0.0004962059,0.0001699772,0.000008984077,0.0006842281],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.007375458,"threshold_uncertainty_score":0.03900564,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1011044871532449,"score_gpt":0.512558837561608,"score_spread":0.4114543504083631,"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."}}