{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0578937,0.001515278,0.003030123,0.003698615,0.001924946,0.007609476,0.003567797,0.003826943,0.002660724],"category_scores_gemma":[0.2475749,0.001341387,0.002656543,0.002415994,0.01202734,0.008094228,0.006655901,0.009529145,0.0005041244],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004260779,"about_ca_system_score_gemma":0.002747722,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003806465,"about_ca_topic_score_gemma":0.002284845,"domain_scores_codex":[0.9557953,0.02844582,0.001826483,0.005980683,0.006898287,0.001053569],"domain_scores_gemma":[0.60867,0.3575191,0.008678254,0.01847531,0.005347679,0.001309655],"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":[0.00007573098,0.00003187875,0.004982019,0.0001919825,0.0002946454,0.0001976871,0.0004378887,0.07256606,0.0003149081,0.9007973,0.001989079,0.01812087],"study_design_scores_gemma":[0.00001373094,0.00003082749,0.0008128652,0.0001040932,0.00002432539,0.00006716891,0.00004039329,0.0953314,0.0001190624,0.9012246,0.002205782,0.00002578394],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02257338,0.005614823,0.950147,0.01110252,0.0003436752,0.00006341729,0.0004235448,0.0003319799,0.009399612],"genre_scores_gemma":[0.791918,0.00712508,0.1861393,0.004105882,0.003210146,0.0008177246,0.001047838,0.0007774219,0.00485859],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0578937,"threshold_uncertainty_score":0,"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."}}