{"id":"W4392622771","doi":"10.2196/53019","title":"Assessing the Quality of ChatGPT Responses to Dementia Caregivers’ Questions: Qualitative Analysis","year":2024,"lang":"en","type":"article","venue":"JMIR Aging","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Institute on Aging; National Institutes of Health; University of Texas at Austin","keywords":"Preprint; Dementia; Quality (philosophy); Psychology; Gerontology; Medicine; Computer science; World Wide Web; Epistemology","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":[0.04713346,0.0004720325,0.0008378205,0.003893593,0.002503814,0.002367215,0.001494607,0.000940973,0.001601959],"category_scores_gemma":[0.1216746,0.0005093718,0.0005932395,0.002792097,0.005359441,0.002370266,0.004618076,0.001300214,0.0002279626],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006108889,"about_ca_system_score_gemma":0.004635076,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00517761,"about_ca_topic_score_gemma":0.00610983,"domain_scores_codex":[0.9672977,0.02484629,0.001824025,0.00102041,0.003545238,0.001466361],"domain_scores_gemma":[0.8519247,0.1191948,0.007790697,0.001794843,0.01676715,0.002528019],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001002154,0.00005120132,0.02919528,0.0006806524,0.00002460057,0.0003188863,0.9519984,0.00008672357,0.001013168,0.0006245855,0.0007649254,0.01514143],"study_design_scores_gemma":[0.00001166945,0.0000874612,0.02017735,0.0005630712,0.00001344928,0.0002320509,0.9731635,0.0003658733,0.0006892076,0.0005133405,0.004142131,0.00004094622],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9901735,0.0003948127,0.003988536,0.001559598,0.00003794787,0.000632317,0.0004896168,0.0000266978,0.002697066],"genre_scores_gemma":[0.9939031,0.0004080917,0.002781498,0.0004085929,0.00001428934,0.001454754,0.0002267217,0.00002437155,0.0007784605],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04713346,"threshold_uncertainty_score":0.2492685,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3026698485249815,"score_gpt":0.6061262897221323,"score_spread":0.3034564411971509,"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."}}