{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001553117,0.0003897302,0.000509059,0.0003184933,0.0001836045,0.0001922619,0.004426708,0.000327257,0.0000286345],"category_scores_gemma":[0.0004568348,0.0003275498,0.0002374051,0.001082505,0.0001364015,0.0003164863,0.002522403,0.001143774,0.0001462186],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001247676,"about_ca_system_score_gemma":0.0003585028,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001266243,"about_ca_topic_score_gemma":0.0002526696,"domain_scores_codex":[0.9967203,0.0003231555,0.000741704,0.001049254,0.0005529678,0.0006126872],"domain_scores_gemma":[0.9963716,0.00054895,0.0003720475,0.002433494,0.0002023515,0.00007161722],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00008659611,0.0007875855,0.3405605,0.0009059568,0.000273646,0.0004734154,0.08137622,0.00712027,0.001695833,0.4679332,0.01210294,0.08668382],"study_design_scores_gemma":[0.0008259034,0.0004517064,0.1411245,0.005596163,0.0001817761,0.00005311712,0.006310767,0.2662946,0.3964727,0.06647579,0.1122844,0.003928623],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7299057,0.0009092276,0.2295722,0.01538798,0.0021906,0.001226073,0.00002744063,0.0002887941,0.02049195],"genre_scores_gemma":[0.9861206,0.0001571723,0.008642329,0.00327962,0.0001933668,0.0001251691,0.000005951462,0.00001860046,0.001457167],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4014574,"threshold_uncertainty_score":0.9999176,"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."}}