{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007516615,0.0008571302,0.0006094545,0.001288083,0.0006030471,0.003917309,0.001473464,0.001340866,0.004739088],"category_scores_gemma":[0.07443975,0.000669126,0.001195502,0.0006091073,0.001869077,0.007411737,0.00255463,0.003083236,0.001300326],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001361608,"about_ca_system_score_gemma":0.001517903,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00361196,"about_ca_topic_score_gemma":0.004633411,"domain_scores_codex":[0.9945082,0.002973497,0.000274605,0.001096172,0.0009000234,0.0002474595],"domain_scores_gemma":[0.939266,0.04030135,0.00527846,0.01201763,0.002319129,0.0008174694],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001798778,0.0004194089,0.09322743,0.001184965,0.001528227,0.001153149,0.005849428,0.1355699,0.01520227,0.2694122,0.02142947,0.4532248],"study_design_scores_gemma":[0.00008122338,0.00008295579,0.006628892,0.00013959,0.0001272978,0.0002406018,0.0004243674,0.496763,0.005797314,0.4835046,0.006125635,0.00008455558],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2010746,0.001060691,0.7739083,0.006360958,0.0001932343,0.0001563919,0.001623456,0.005488127,0.01013424],"genre_scores_gemma":[0.9362258,0.0002355937,0.05843314,0.0008856026,0.0001075593,0.00008825491,0.001719098,0.0006231138,0.001681648],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007516615,"threshold_uncertainty_score":0.03975207,"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."}}