{"id":"W2057277282","doi":"10.1119/1.4849159","title":"Physics Exams that Promote Collaborative Learning","year":2013,"lang":"en","type":"article","venue":"The Physics Teacher","topic":"Experimental Learning in Engineering","field":"Engineering","cited_by":65,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Formative assessment; Summative assessment; Mathematics education; Consistency (knowledge bases); Physics education; Collaborative learning; Computer science; Medical education; Student engagement; Psychology; Medicine; Artificial intelligence","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":[0.00009766728,0.0002326452,0.0001677923,0.00001220145,0.0001241001,0.00009338572,0.0002365967,0.00005269086,0.0001157668],"category_scores_gemma":[0.00001149134,0.0001857972,0.00005614006,0.0002449818,0.00005916719,0.000297196,0.0000653408,0.0006142948,0.000697206],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000911978,"about_ca_system_score_gemma":0.000007968199,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002227496,"about_ca_topic_score_gemma":3.224672e-7,"domain_scores_codex":[0.9991553,0.00005381695,0.0001125321,0.0001581548,0.0001934441,0.0003267124],"domain_scores_gemma":[0.9995387,0.00004811764,0.00003506494,0.0002916994,0.00003055462,0.00005584549],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000002649479,0.00007725645,0.001549446,0.00004713452,0.0002202075,0.000001796859,0.01712373,0.7404764,0.1848283,0.00181968,0.002503803,0.05134965],"study_design_scores_gemma":[0.0005238678,0.00009233614,0.002667267,0.00009198573,0.00004763358,0.000003977844,0.002795791,0.7001947,0.2713569,0.003058775,0.0181883,0.0009785446],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9219204,0.0004877689,0.03005757,0.00009920966,0.0003709421,0.0006305412,0.000002007761,0.00153414,0.04489741],"genre_scores_gemma":[0.9967018,0.00001451722,0.001074593,0.00002008526,0.0004094727,0.0001259436,0.000008286401,0.0001123971,0.001532873],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.08652858,"threshold_uncertainty_score":0.8961406,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01094219117019928,"score_gpt":0.2093543674537619,"score_spread":0.1984121762835626,"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."}}