{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002152054,0.000343676,0.0005613817,0.0001542525,0.001060396,0.00002794065,0.0005531118,0.0001108783,0.0006636445],"category_scores_gemma":[0.00002220031,0.0002785387,0.0001310899,0.0009228296,0.0001457421,0.0001715457,0.0003228453,0.0005704064,0.00100748],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005053702,"about_ca_system_score_gemma":0.0001600518,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01295341,"about_ca_topic_score_gemma":0.001827016,"domain_scores_codex":[0.9948275,0.001242097,0.001079937,0.0006871831,0.0008360832,0.001327218],"domain_scores_gemma":[0.9975245,0.0005004915,0.0007828883,0.00091069,0.00004368617,0.0002377455],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005366931,0.007188769,0.3497353,0.001194568,0.0009948364,0.0001274203,0.324754,0.0000199749,0.0002139579,0.006888638,0.1070644,0.1964512],"study_design_scores_gemma":[0.01162131,0.00874037,0.6688417,0.0003206276,0.00005069411,0.00001917939,0.295716,0.0001655582,0.0001298133,0.0003803715,0.01349446,0.0005199304],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9796874,0.0007423693,0.00001028694,0.003136259,0.003318722,0.004736539,0.0002697746,0.0002613567,0.007837293],"genre_scores_gemma":[0.9955645,0.0002428289,0.0001428482,0.001357016,0.0002970103,0.0008262749,0.0001267214,0.00007866934,0.001364182],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3191064,"threshold_uncertainty_score":0.9999667,"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."}}