{"id":"W7082288996","doi":"10.48448/hdgp-ad50","title":"Hallucination Detox: Sensitivity Dropout (SenD) for Large Language Model Training","year":2025,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"Geochemistry and Geologic Mapping","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Mila - Quebec Artificial Intelligence Institute; Polytechnique Montréal","funders":"","keywords":"Sensitivity (control systems); Task (project management); Embedding; Dropout (neural networks); Reliability (semiconductor); Auditory hallucination; Variance (accounting); Language model; Metric (unit)","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":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001638514,0.0002733667,0.0003063031,0.0003895626,0.0003340873,0.0001933005,0.001024782,0.0002334141,0.00005595161],"category_scores_gemma":[0.0006902518,0.0002622352,0.00007763771,0.0006452163,0.0001844531,0.0002423092,0.0004767536,0.0002367849,0.00001854402],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008829469,"about_ca_system_score_gemma":0.0009606695,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005886998,"about_ca_topic_score_gemma":0.0002667466,"domain_scores_codex":[0.9976914,0.00003974366,0.0002321734,0.0009393775,0.0004510157,0.0006463108],"domain_scores_gemma":[0.9985633,0.0001421196,0.0002148922,0.0007364987,0.0002311779,0.0001119861],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000009825123,0.0003412427,0.00003789267,0.0005905132,0.00008222563,0.00009879877,0.005299027,0.009906096,0.01322318,0.6673834,0.09581402,0.2072138],"study_design_scores_gemma":[0.000255802,0.00001749578,0.000003643162,0.0001096335,0.00001130529,0.00000890446,0.0001981317,0.9517232,0.0009353208,0.00557042,0.04087297,0.0002931727],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"other","genre_scores_codex":[0.00001044543,0.00005874317,0.6794919,0.001059912,0.0002809209,0.0002263464,0.00004758924,0.000296937,0.3185272],"genre_scores_gemma":[0.1530698,0.00001010169,0.2726502,0.0007379159,0.0002539156,0.00002826375,0.00006505472,0.0000302856,0.5731545],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9418171,"threshold_uncertainty_score":0.999983,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02421322823962332,"score_gpt":0.2906454926543322,"score_spread":0.2664322644147089,"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."}}