{"id":"W4417334214","doi":"10.1177/18344909251406111","title":"Evaluating the Critical Thinking of Large Language Models: Insights and Limitations","year":2025,"lang":"en","type":"article","venue":"Journal of Pacific Rim Psychology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"National Natural Science Foundation of China","keywords":"Critical thinking; Argumentative; Cognition; Project commissioning; Key (lock); Publishing; Professional writing; Writing assessment","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001000134,0.00005345404,0.0001824677,0.0001693998,0.0001249792,0.000009647614,0.00007253868,0.00008084805,0.0000189431],"category_scores_gemma":[0.001431713,0.00003437193,0.00005432725,0.0001792168,0.0001207355,0.00007803847,0.00001182695,0.0003031869,0.000002119066],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001982389,"about_ca_system_score_gemma":0.0001677059,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001016234,"about_ca_topic_score_gemma":0.00001333468,"domain_scores_codex":[0.9989738,0.000157295,0.0005091474,0.00008705492,0.0001555294,0.0001171328],"domain_scores_gemma":[0.9981893,0.001059531,0.0001590427,0.0001540571,0.0003909,0.00004718014],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001354696,0.001558786,0.01097416,0.0005237212,0.0003502138,0.00007451785,0.267203,0.0001995541,0.04183292,0.2208948,0.0129528,0.4420809],"study_design_scores_gemma":[0.0008566321,0.003885278,0.02150522,0.001340384,0.000637147,0.0008482565,0.2485972,0.01515514,0.005299753,0.696769,0.004895361,0.000210563],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.942457,0.005180649,0.01196197,0.02601211,0.001142676,0.0001261072,9.632047e-7,0.000005757483,0.01311276],"genre_scores_gemma":[0.9952963,0.0004480751,0.002730173,0.001233981,0.0001739781,0.000002550406,9.265036e-7,0.000004241183,0.0001097925],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4758742,"threshold_uncertainty_score":0.1713998,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3430885522212014,"score_gpt":0.5717520442325003,"score_spread":0.2286634920112989,"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."}}