{"id":"W4407950712","doi":"10.1109/fie61694.2024.10893214","title":"WIP: Assessing the Effectiveness of ChatGPT in Preparatory Testing Activities","year":2024,"lang":"en","type":"article","venue":"","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University; Fanshawe College","funders":"","keywords":"Computer science; Reliability engineering; Engineering","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":[0.01616738,0.001045751,0.0008634544,0.002367189,0.0007398704,0.001818105,0.002016446,0.001442989,0.002137615],"category_scores_gemma":[0.1009685,0.0006280809,0.0006044409,0.0008741349,0.0008488832,0.002674878,0.00244559,0.001160723,0.001105756],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007594718,"about_ca_system_score_gemma":0.001251202,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001064657,"about_ca_topic_score_gemma":0.001578582,"domain_scores_codex":[0.9823346,0.009132422,0.00201154,0.001997577,0.003801958,0.0007219498],"domain_scores_gemma":[0.8661765,0.09974889,0.01343231,0.007219417,0.009037277,0.004385529],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.004746556,0.01797494,0.1464682,0.002450639,0.0003068939,0.0006304897,0.01835537,0.004122894,0.03698222,0.0006316513,0.002747363,0.7645829],"study_design_scores_gemma":[0.002193734,0.152027,0.6722618,0.001593325,0.001365606,0.002765644,0.01893108,0.03981493,0.08079814,0.002234344,0.0253972,0.0006171795],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.970912,0.0002713238,0.02039062,0.0002428459,0.00005650903,0.001943519,0.0001669812,0.0007166675,0.005299594],"genre_scores_gemma":[0.9065089,0.0003174893,0.08742722,0.0002823391,0.00007265028,0.002420246,0.0003732136,0.0001068409,0.002491159],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01616738,"threshold_uncertainty_score":0.08550233,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.168787798421792,"score_gpt":0.4861090872566501,"score_spread":0.3173212888348581,"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."}}