{"id":"W4415169429","doi":"10.1007/s10639-025-13801-4","title":"Generative AI in Data Science Programming: Differences in Performance Between Groups With and Without AI-assistance","year":2025,"lang":"en","type":"article","venue":"Education and Information Technologies","topic":"Innovative Human-Technology Interaction","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"Simon Fraser University","funders":"British Columbia Knowledge Development Fund; Canada Foundation for Innovation","keywords":"Context (archaeology); Generative grammar; Science education; Significant difference; Educational technology","routes":{"ca_aff":true,"ca_fund":true,"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":[0.0003786955,0.0001214363,0.0001446696,0.001400513,0.0001975874,0.0003559767,0.0008097412,0.0000942298,4.575504e-7],"category_scores_gemma":[0.0001879296,0.0000986115,0.000003185593,0.002122897,0.0006197408,0.01114613,0.0004410983,0.0003362693,0.000001834206],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00011189,"about_ca_system_score_gemma":0.0003378394,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001610194,"about_ca_topic_score_gemma":0.00005156096,"domain_scores_codex":[0.9990563,0.00001457348,0.0003050184,0.0002894947,0.0001466231,0.000187937],"domain_scores_gemma":[0.9991385,0.00002446133,0.0001510975,0.0004707906,0.0002036653,0.00001153287],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000005753647,0.00003223372,0.4352972,0.00003603298,0.000003657493,5.342378e-8,0.001087543,8.392477e-7,0.00004568033,0.1456289,0.00009069699,0.4177715],"study_design_scores_gemma":[0.000399568,0.0001056689,0.9641051,0.0002674426,0.000003232397,0.000006933176,0.006114641,0.01275549,0.005596915,0.004890252,0.005525049,0.0002297478],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9375926,0.0001017811,0.04749451,0.01325699,0.0001441914,0.0003593926,0.000001546949,0.0003158029,0.000733205],"genre_scores_gemma":[0.98413,0.00009881254,0.01521528,0.000405674,0.000004160341,0.0001014751,0.00001205696,0.000001467505,0.00003109992],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5288079,"threshold_uncertainty_score":0.8080673,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02383347926566382,"score_gpt":0.3105078002408833,"score_spread":0.2866743209752194,"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."}}