{"id":"W4417091950","doi":"10.48550/arxiv.2504.13151","title":"MIB: A Mechanistic Interpretability Benchmark","year":2025,"lang":"en","type":"preprint","venue":"UvA-DARE (University of Amsterdam)","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Azrieli Foundation; European Commission; Open Philanthropy Project","keywords":"Interpretability; Benchmark (surveying); Task (project management); Variable (mathematics); Causal model; Artificial neural network; Task analysis","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":[0.0005969585,0.0004264625,0.0007476312,0.0004880067,0.0003069772,0.0001505367,0.004617218,0.0004319024,0.0004101995],"category_scores_gemma":[0.0001683477,0.0005642209,0.0004671121,0.0006328622,0.0002898451,0.0005546522,0.00756671,0.0007628327,0.0001238028],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004029673,"about_ca_system_score_gemma":0.000640279,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002192857,"about_ca_topic_score_gemma":0.0009289586,"domain_scores_codex":[0.9969385,0.0002924035,0.0004102373,0.001311087,0.0005301231,0.0005176822],"domain_scores_gemma":[0.996144,0.0003064106,0.0004457715,0.002344469,0.00056074,0.0001986377],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0006298849,0.002013864,0.002791082,0.007120354,0.001438262,0.001194718,0.08054011,0.005328518,0.002777314,0.5775725,0.01814429,0.3004491],"study_design_scores_gemma":[0.00110677,0.0009396511,0.004016167,0.004698787,0.0006060455,0.00003489984,0.01439626,0.4352851,0.01155705,0.4908344,0.03264484,0.003880046],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02595531,0.000191441,0.9370241,0.001693761,0.001775978,0.0007479501,0.000184252,0.0003657186,0.03206151],"genre_scores_gemma":[0.9391177,0.00006233674,0.05603364,0.000206718,0.00005185293,0.000002681687,0.00004587659,0.00001514405,0.004463994],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9131625,"threshold_uncertainty_score":0.9996809,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01612328950733706,"score_gpt":0.2312647162251226,"score_spread":0.2151414267177855,"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."}}