{"id":"W4410979322","doi":"10.53103/cjess.v5i3.364","title":"ChatGPT: Hero or Villain? Comparative Evaluation by Canadian High School Students and Teachers","year":2025,"lang":"en","type":"article","venue":"Canadian Journal of Educational and Social Studies","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Barrie Urology Group","funders":"","keywords":"HERO; Mathematics education; Pedagogy; Psychology; Sociology; Art; Literature","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.01530722,0.0006557085,0.001184506,0.002825564,0.004828463,0.002603611,0.001485309,0.0009646004,0.004677154],"category_scores_gemma":[0.02342127,0.0005298647,0.0009192908,0.003042163,0.002236059,0.001569285,0.001868431,0.001076093,0.0006254424],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01889157,"about_ca_system_score_gemma":0.03191451,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.64059,"about_ca_topic_score_gemma":0.8263369,"domain_scores_codex":[0.989549,0.00384078,0.0006206309,0.000675683,0.003730444,0.001583418],"domain_scores_gemma":[0.9680657,0.005475685,0.001723685,0.0008887451,0.01629658,0.007549598],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.008904302,0.02333437,0.2655994,0.005392869,0.0002907888,0.001289706,0.3396741,0.0005534761,0.004716305,0.001354959,0.01594217,0.3329476],"study_design_scores_gemma":[0.001218726,0.01406238,0.7090039,0.001516683,0.0003734927,0.0002814444,0.2381807,0.0009270167,0.002051881,0.0001914975,0.0319769,0.0002155032],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9915273,0.0005837975,0.0002018832,0.0002472994,0.0000469364,0.00138628,0.0004193874,0.00004087628,0.005546369],"genre_scores_gemma":[0.9907842,0.001277778,0.001344917,0.0002346797,0.0000400769,0.001518755,0.0008395577,0.00003157138,0.003928604],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9811084,"threshold_uncertainty_score":0.7230532,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2550967466908575,"score_gpt":0.5121983646571294,"score_spread":0.257101617966272,"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."}}