{"id":"W4388212323","doi":"10.1109/issre59848.2023.00052","title":"Resilience Assessment of Large Language Models under Transient Hardware Faults","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Computer science; Transient (computer programming); Reliability (semiconductor); Resilience (materials science); Power (physics)","routes":{"ca_aff":true,"ca_fund":true,"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":[0.002356046,0.001136447,0.0006786502,0.0008572327,0.0005512778,0.0008787999,0.001153865,0.001209473,0.001371124],"category_scores_gemma":[0.02285613,0.0004713341,0.0008710842,0.0003984075,0.001087682,0.001635507,0.001228135,0.001683471,0.0004083583],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001191282,"about_ca_system_score_gemma":0.001124003,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008248251,"about_ca_topic_score_gemma":0.005122399,"domain_scores_codex":[0.9983763,0.0004964454,0.0001724493,0.0003195321,0.0004117518,0.0002234594],"domain_scores_gemma":[0.9851638,0.01035404,0.001090799,0.001941768,0.001070541,0.0003789799],"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":[0.0006667473,0.00009650402,0.007439842,0.00023949,0.0001474567,0.000317629,0.000214051,0.9563713,0.01080423,0.00146664,0.001174668,0.02106149],"study_design_scores_gemma":[0.00001746189,0.0001594743,0.00110138,0.00001635338,0.00003102726,0.00006019394,0.00007161315,0.9882918,0.008599219,0.001319166,0.000317791,0.00001450206],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8872518,0.001120567,0.09993163,0.0008730859,0.000179543,0.0001269284,0.0008916603,0.007178566,0.002446136],"genre_scores_gemma":[0.9905861,0.00009143921,0.008260002,0.00007455261,0.00001085479,0.00004313268,0.0004544089,0.0001504609,0.0003291317],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008248251,"threshold_uncertainty_score":0.01640046,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03433932826595434,"score_gpt":0.3259301749395708,"score_spread":0.2915908466736165,"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."}}