{"id":"W7126397965","doi":"10.21428/594757db.e9f456d0","title":"Generative AI for Enhancing Active Learning in Education:A Comparative Study of GPT-3.5 and GPT-4 in CraftingCustomized Test Questions","year":2024,"lang":"en","type":"article","venue":"","topic":"Intelligent Tutoring Systems and Adaptive Learning","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ontario Tech University","funders":"","keywords":"Active learning (machine learning); Test (biology); Generative grammar; Path (computing); Generative model","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":[0.005001027,0.0006266796,0.0006128469,0.0007667994,0.0003582949,0.001658322,0.001949285,0.0009277345,0.002718445],"category_scores_gemma":[0.02106966,0.0002931193,0.0005050183,0.0005422387,0.001149579,0.001954443,0.002159787,0.001501121,0.000686146],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001181966,"about_ca_system_score_gemma":0.0008574667,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001056395,"about_ca_topic_score_gemma":0.001371239,"domain_scores_codex":[0.9970639,0.001671996,0.0002039886,0.0003418303,0.0005266209,0.0001915921],"domain_scores_gemma":[0.9876023,0.009658479,0.0004823054,0.001087663,0.0004152834,0.0007539261],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.004696324,0.0308263,0.03441796,0.003319508,0.0003536359,0.0009038259,0.0789587,0.03039286,0.05454938,0.01363765,0.003892769,0.7440511],"study_design_scores_gemma":[0.006096866,0.1220323,0.2526164,0.001645533,0.001350252,0.003961775,0.04404435,0.2251618,0.130879,0.04315547,0.1681209,0.0009353244],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9838973,0.0001329721,0.009818476,0.0001337557,0.00001764811,0.0004510552,0.00007647341,0.0002332967,0.005239043],"genre_scores_gemma":[0.9651447,0.000254652,0.03057255,0.0001821757,0.000011145,0.0007075866,0.0002372841,0.00006744699,0.002822541],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005001027,"threshold_uncertainty_score":0.02644825,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02652257507694267,"score_gpt":0.3367431233355652,"score_spread":0.3102205482586225,"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."}}