{"id":"W7126419163","doi":"10.21428/594757db.ef83c701","title":"Does ChatGPT Measure Up to Discourse Unit Segmentation?A Comparative Analysis Utilizing Zero-Shot Custom Prompts","year":2024,"lang":"en","type":"article","venue":"","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Measure (data warehouse); Unit (ring theory); Hallucinating; Segmentation; Natural language; Usability; Natural (archaeology)","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":[0.01130003,0.0009453817,0.0009427866,0.002691458,0.0006087616,0.002313295,0.001235768,0.001403567,0.002331272],"category_scores_gemma":[0.1221742,0.0004149013,0.0004859336,0.001364158,0.001338696,0.003800192,0.003183524,0.001325935,0.001107212],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000946795,"about_ca_system_score_gemma":0.0005768352,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002388477,"about_ca_topic_score_gemma":0.002367637,"domain_scores_codex":[0.9893489,0.007002422,0.000559266,0.001437235,0.001348348,0.0003037555],"domain_scores_gemma":[0.8368878,0.1381931,0.00482426,0.007438799,0.01070571,0.001950416],"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":[0.009244613,0.0009406216,0.2087887,0.006000455,0.0007885735,0.002146079,0.05943953,0.01348014,0.06775089,0.00385532,0.007674103,0.619891],"study_design_scores_gemma":[0.000310599,0.008160462,0.7305405,0.001049751,0.0008313233,0.003215982,0.03267583,0.1328315,0.05990212,0.007942398,0.02200812,0.0005314026],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9559109,0.001105226,0.03062219,0.0002779718,0.0001675539,0.0003216226,0.001450067,0.002829218,0.007315078],"genre_scores_gemma":[0.9857927,0.0001862366,0.01062126,0.0000912043,0.00004577755,0.0001745944,0.001343498,0.0003110413,0.001433728],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01130003,"threshold_uncertainty_score":0.05976099,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3010559395340812,"score_gpt":0.5138741071835894,"score_spread":0.2128181676495082,"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."}}