{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005118363,0.001555604,0.0008820733,0.0006706087,0.0007190381,0.001169535,0.002351852,0.001621152,0.003531794],"category_scores_gemma":[0.03105498,0.0007778669,0.0008171518,0.0005050241,0.001096406,0.002572051,0.003606817,0.003494311,0.001460319],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001012405,"about_ca_system_score_gemma":0.001674056,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003227954,"about_ca_topic_score_gemma":0.007314856,"domain_scores_codex":[0.9975173,0.001426197,0.0001663138,0.0003784508,0.0003588217,0.0001527935],"domain_scores_gemma":[0.9881258,0.008244536,0.0004739395,0.001898229,0.0009349539,0.0003226551],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001338201,0.0005972476,0.007410065,0.0004353829,0.0003883899,0.0004331944,0.0006074713,0.5359734,0.01374716,0.01039733,0.02135273,0.4073194],"study_design_scores_gemma":[0.00003845384,0.0001144557,0.0002036023,0.00001240832,0.00001226094,0.00003996008,0.00002751544,0.9892302,0.004382794,0.005333982,0.0005919693,0.00001238986],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.08170496,0.0005606411,0.8929639,0.001076039,0.0001888247,0.0002516786,0.0005506232,0.0203986,0.002304595],"genre_scores_gemma":[0.6763818,0.0001677295,0.3162566,0.0008498493,0.0000946197,0.0005616266,0.001464378,0.0013691,0.002854453],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.005118363,"threshold_uncertainty_score":0.02706879,"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."}}