{"id":"W6966817453","doi":"10.48448/4z73-8143","title":"Are self-explanations from Large Language Models faithful?","year":2024,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Mila - Quebec Artificial Intelligence Institute; Polytechnique Montréal","funders":"","keywords":"Measure (data warehouse); Feature (linguistics); Counterfactual conditional; Set (abstract data type); Language model; Redaction; Counterfactual thinking; Inference","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","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0007544396,0.0005554743,0.0004920423,0.00194683,0.0002204543,0.0004862999,0.001613841,0.0003479469,0.004010895],"category_scores_gemma":[0.0001465723,0.0004957144,0.0001154933,0.002336532,0.0005053727,0.0004613852,0.0005298161,0.0006471597,0.04533491],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000490665,"about_ca_system_score_gemma":0.0006088233,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001420828,"about_ca_topic_score_gemma":0.00871239,"domain_scores_codex":[0.9955214,0.00006453222,0.0003824712,0.001468238,0.001639075,0.0009242579],"domain_scores_gemma":[0.9977108,0.00005846101,0.0004858146,0.001263861,0.0001545636,0.0003265633],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000002170842,0.0002240038,0.00004875415,0.00005314482,0.0001117948,0.0001551466,0.001404885,0.0001274835,0.0004339864,0.01424934,0.9829915,0.0001977669],"study_design_scores_gemma":[0.000734141,0.00003271169,0.00006124325,0.0008231772,0.0003305791,0.00001712136,0.004734345,0.1896363,0.0001604318,0.01456778,0.7874067,0.001495449],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.0003787467,0.004869409,0.005058952,0.0004357625,0.001604484,0.0007769217,0.01830512,0.005285369,0.9632852],"genre_scores_gemma":[0.08451772,0.00008316595,0.03234678,0.001044994,0.002265691,0.0001152782,0.00166211,0.003400364,0.8745639],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.1955848,"threshold_uncertainty_score":0.9997494,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02546910737271367,"score_gpt":0.3056888415430832,"score_spread":0.2802197341703695,"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."}}