{"id":"W4416036894","doi":"10.18653/v1/2025.emnlp-main.519","title":"SAEs Are Good for Steering – If You Select the Right Features","year":2025,"lang":"","type":"article","venue":"","topic":"Vehicle Dynamics and Control Systems","field":"Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Azrieli Foundation; European Commission; Open Philanthropy Project","keywords":"Set (abstract data type); Feature (linguistics); Matching (statistics); Identification (biology); Key (lock)","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":[0.0003988464,0.000419107,0.0005264619,0.0001182257,0.0004384709,0.0004320283,0.0005440206,0.0002585011,0.00006970268],"category_scores_gemma":[0.00004716998,0.0002943327,0.0003427078,0.0004084496,0.00003338542,0.0001077582,0.00008623573,0.0003682761,0.00001481174],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001571768,"about_ca_system_score_gemma":0.00006150164,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001613201,"about_ca_topic_score_gemma":0.0008138188,"domain_scores_codex":[0.9981461,0.00005085119,0.000497864,0.0004132989,0.0001979831,0.0006939535],"domain_scores_gemma":[0.9988134,0.0003108813,0.00009217097,0.0005733618,0.0001309476,0.00007924293],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001923,0.0001771386,0.00381236,0.00295769,0.004290737,0.00002058339,0.001179757,0.2041845,0.009101169,0.6611171,0.0311671,0.08179951],"study_design_scores_gemma":[0.001353909,0.00007012828,0.007716987,0.0003830615,0.0002484051,0.000009324102,0.0007337716,0.847488,0.0008593651,0.001033636,0.1396024,0.0005010423],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.08771769,0.05280101,0.3610454,0.02004109,0.01621274,0.009541452,0.000563735,0.001802071,0.4502747],"genre_scores_gemma":[0.9580072,0.0001306665,0.0001627024,0.0002047683,0.000410285,0.0002288353,0.000007244008,0.00005807544,0.04079021],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8702895,"threshold_uncertainty_score":0.9999509,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.002906539219855812,"score_gpt":0.1958968182108761,"score_spread":0.1929902789910203,"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."}}