{"id":"W4416030400","doi":"10.48550/arxiv.2503.08679","title":"Chain-of-Thought Reasoning In The Wild Is Not Always Faithful","year":2025,"lang":"en","type":"preprint","venue":"ArXiv.org","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Institute for Catastrophic Loss Reduction","keywords":"Rationalization (economics); Confabulation (neural networks); Sonnet; Face (sociological concept); Analogical reasoning; Psychology of reasoning","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":[0.01783148,0.001129673,0.000566003,0.0007380142,0.0009639707,0.005122856,0.0017082,0.001870645,0.01106564],"category_scores_gemma":[0.1457399,0.0007071485,0.0008705267,0.0005373896,0.003892544,0.009104358,0.003636032,0.003394592,0.002658382],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002076759,"about_ca_system_score_gemma":0.002198366,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003206268,"about_ca_topic_score_gemma":0.00210378,"domain_scores_codex":[0.9726416,0.01736362,0.001206905,0.003630124,0.004332273,0.0008254016],"domain_scores_gemma":[0.8524325,0.1054875,0.01107068,0.02315564,0.006567551,0.001286215],"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":[0.002708454,0.0005003134,0.1085659,0.002636621,0.0005910958,0.001707635,0.03883797,0.09253734,0.03523454,0.3350737,0.03055741,0.3510492],"study_design_scores_gemma":[0.0002308381,0.0004915408,0.01140787,0.0007839992,0.000207085,0.0009008067,0.005764192,0.2535643,0.02687519,0.6308537,0.06866474,0.0002557864],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3704147,0.0007068329,0.5563247,0.006170262,0.0003816803,0.0003500609,0.001257427,0.00937915,0.05501512],"genre_scores_gemma":[0.9604642,0.0001397014,0.03503713,0.0005883331,0.00003765452,0.00008249137,0.0005600267,0.0005855101,0.002505002],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01783148,"threshold_uncertainty_score":0.09430295,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0654847306672032,"score_gpt":0.3068589928171223,"score_spread":0.2413742621499191,"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."}}