{"id":"W4390905989","doi":"10.1109/transai60598.2023.00047","title":"On the Variability of Statistical Models","year":2023,"lang":"en","type":"article","venue":"","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Computer science","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"grok","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"},{"model":"opus","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001247312,0.00005018758,0.0001204354,0.00001435374,0.00002695137,0.000005822022,0.00009860488,0.00002538692,0.001703643],"category_scores_gemma":[0.01023158,0.00002567425,0.0000206977,0.0001312045,0.00008267837,0.000010261,0.00003839505,0.00007324847,0.00006218296],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000005918104,"about_ca_system_score_gemma":0.00001610708,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001129437,"about_ca_topic_score_gemma":7.417844e-7,"domain_scores_codex":[0.9992544,0.0001937395,0.0001767026,0.00009933325,0.0001671334,0.0001086668],"domain_scores_gemma":[0.9771861,0.0224656,0.00002519014,0.0002491588,0.0000441461,0.00002975966],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000005571864,0.00003240363,0.00001631666,0.00002009511,0.000004870237,5.514763e-7,0.00004642366,0.000009662639,0.00003035744,0.989911,0.007911118,0.002011624],"study_design_scores_gemma":[0.00003633642,0.00003831939,0.0004880305,0.000006341792,0.000005406016,1.333406e-7,0.00002804261,0.08699092,0.0001657383,0.9121949,0.000015038,0.00003080809],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.0170683,1.794068e-7,0.9015325,0.0003086291,0.00003553092,0.00009445113,0.00005481886,0.00004920217,0.0808564],"genre_scores_gemma":[0.7037379,0.000001020555,0.2958055,0.00009509626,0.000007967623,0.0000113533,9.270079e-7,0.000005506145,0.0003347419],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6866696,"threshold_uncertainty_score":0.9992089,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2362420914277805,"score_gpt":0.4272419507385115,"score_spread":0.1909998593107309,"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."}}