{"id":"W7101411521","doi":"10.1063/4.0000863","title":": “I’m sorry, Dave, I’m afraid I can’t do that” Part 2","year":2025,"lang":"en","type":"article","venue":"Structural Dynamics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Session (web analytics); Software; Text recognition; Text detection","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":[0.00009143441,0.0001443838,0.0002056002,0.0001120488,0.000173938,0.00003724816,0.0001089243,0.0001166698,0.0003540636],"category_scores_gemma":[0.000121987,0.000123012,0.00007655497,0.0002694945,0.00009896814,0.00006387707,0.00003746523,0.0002643171,0.00002698863],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002348095,"about_ca_system_score_gemma":0.0003400254,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000696689,"about_ca_topic_score_gemma":0.00143784,"domain_scores_codex":[0.9989459,0.00002628303,0.0002850068,0.0002586359,0.0001711529,0.000313031],"domain_scores_gemma":[0.9992727,0.00006997836,0.00006227279,0.000338349,0.0001388358,0.0001178819],"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":[0.0001566997,0.00003258528,0.3159528,0.0003068274,0.00007911173,0.00001804166,0.001654767,0.00004436901,0.0001168977,0.05246351,0.008642853,0.6205315],"study_design_scores_gemma":[0.0005496376,0.0006854319,0.490687,0.00114068,0.0005200995,0.0002303147,0.02320087,0.06571683,0.009014136,0.3431128,0.06380131,0.001340888],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9700734,0.0002322724,0.0002819691,0.01070814,0.003296018,0.0003869789,0.00003393693,0.00009833912,0.01488894],"genre_scores_gemma":[0.9910626,0.00007484066,0.0004019569,0.001893118,0.0002896138,0.00001581761,0.0001854252,0.00001313485,0.006063487],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6191906,"threshold_uncertainty_score":0.5016285,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06922315228168077,"score_gpt":0.4017432386545842,"score_spread":0.3325200863729034,"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."}}