{"id":"W4402683840","doi":"10.18653/v1/2024.findings-acl.19","title":"Are self-explanations from Large Language Models faithful?","year":2024,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Alliance de recherche numérique du Canada; Natural Sciences and Engineering Research Council of Canada; Canadian Institute for Advanced Research","keywords":"Computer science; Natural language processing; Language model; Artificial intelligence; Linguistics; Philosophy","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01164947,0.0009817661,0.0008896022,0.001726886,0.0008189693,0.004329748,0.001993801,0.001706688,0.003588255],"category_scores_gemma":[0.1111797,0.001034739,0.001596327,0.0007458307,0.002620952,0.009400059,0.003163153,0.004130177,0.0008349777],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001543889,"about_ca_system_score_gemma":0.001787623,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003541689,"about_ca_topic_score_gemma":0.004498467,"domain_scores_codex":[0.9915558,0.0045281,0.0004949336,0.001556776,0.001457178,0.0004071458],"domain_scores_gemma":[0.8850297,0.07908924,0.00898893,0.02185809,0.00387952,0.001154591],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002423025,0.0005465733,0.1114117,0.001411691,0.002187257,0.001386511,0.008805436,0.1885186,0.01542209,0.281253,0.01412306,0.3725111],"study_design_scores_gemma":[0.00009821729,0.0001119949,0.006312781,0.0001549014,0.0001618088,0.0002715576,0.0004988066,0.5129018,0.005872414,0.4697159,0.003805631,0.00009420346],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2164728,0.0008743557,0.7680032,0.004664151,0.0001606559,0.0001811602,0.001046524,0.003579322,0.005017843],"genre_scores_gemma":[0.9402697,0.0001953257,0.05596534,0.0007753348,0.0001247195,0.00009694121,0.001216775,0.000435632,0.0009201806],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01164947,"threshold_uncertainty_score":0.06160903,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02192689832023895,"score_gpt":0.2564935360003454,"score_spread":0.2345666376801065,"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."}}