{"id":"W4416036789","doi":"10.18653/v1/2025.emnlp-main.504","title":"Measuring Chain of Thought Faithfulness by Unlearning Reasoning Steps","year":2025,"lang":"","type":"article","venue":"","topic":"Mind wandering and attention","field":"Neuroscience","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"Azrieli Foundation; European Commission; Open Philanthropy Project","keywords":"Chain (unit); Identification (biology); Measure (data warehouse)","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":[0.002664598,0.0004961388,0.0003855448,0.001088255,0.0003184061,0.002007938,0.0005456338,0.0007395849,0.004831005],"category_scores_gemma":[0.05623526,0.000445642,0.0005200442,0.0005312914,0.0006628418,0.002088472,0.0009666525,0.001626641,0.0006601269],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004398383,"about_ca_system_score_gemma":0.0006813192,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001070303,"about_ca_topic_score_gemma":0.001456911,"domain_scores_codex":[0.9981129,0.000348835,0.0001849836,0.0005415016,0.0007038718,0.0001078184],"domain_scores_gemma":[0.9595411,0.02925893,0.003758256,0.004039491,0.002465042,0.0009371894],"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":[0.006064774,0.002175036,0.327548,0.0006124085,0.0009637502,0.0002829388,0.003974234,0.01960862,0.1251098,0.01329462,0.00113058,0.4992352],"study_design_scores_gemma":[0.0004251797,0.00434114,0.6316763,0.000263459,0.0007018041,0.001099145,0.001172519,0.2077923,0.09064487,0.05791912,0.00370965,0.0002544929],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8649625,0.0003435022,0.1268625,0.0001289381,0.00007187133,0.0004266886,0.0003789612,0.0008420831,0.005982815],"genre_scores_gemma":[0.9569952,0.0001185933,0.04118969,0.00005301768,0.00001465335,0.0001336418,0.0003697244,0.00007459812,0.001050871],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004831005,"threshold_uncertainty_score":0.01616138,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02972615203049005,"score_gpt":0.2586175099565577,"score_spread":0.2288913579260677,"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."}}