{"id":"W4389518693","doi":"10.18653/v1/2023.emnlp-main.403","title":"Understanding the Inner-workings of Language Models Through Representation Dissimilarity","year":2023,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Thomson Reuters (Canada)","funders":"Pacific Northwest National Laboratory; Laboratory Directed Research and Development; Battelle; U.S. Department of Energy","keywords":"Interpretability; Computer science; Representation (politics); Language model; Generalization; Measure (data warehouse); Set (abstract data type); Transparency (behavior); Natural language processing; Artificial intelligence; Machine learning; Data mining; Mathematics; Programming language","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002826391,0.0000527245,0.0000746373,0.00003737665,0.00009229984,0.00005741779,0.000461387,0.00002822832,0.000007778051],"category_scores_gemma":[0.00002777827,0.00003561269,0.00003680148,0.000556752,0.00002755758,0.0004455996,0.0002534343,0.00007270977,0.000004919493],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002999564,"about_ca_system_score_gemma":0.00001652007,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002070674,"about_ca_topic_score_gemma":0.00002336338,"domain_scores_codex":[0.999253,0.00004058873,0.0001494288,0.0001917889,0.0002280175,0.0001371159],"domain_scores_gemma":[0.9993039,0.000135944,0.00005253238,0.0004741244,0.00001802874,0.00001544279],"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":[0.000001165861,0.00000653193,0.000319717,0.000007754069,0.000008265138,0.000003728664,0.01732339,0.02932327,0.00036306,0.9498709,0.0008822891,0.001889962],"study_design_scores_gemma":[0.00005949058,0.000003786697,0.0001291073,0.00000855847,0.00000178515,8.970273e-7,0.002144187,0.7448225,0.0009215074,0.2518497,0.00001835139,0.00004005956],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02027275,0.00001674467,0.9598176,0.00292161,0.0001314124,0.00008090074,5.187407e-7,0.0001962219,0.0165622],"genre_scores_gemma":[0.984224,0.00001117375,0.01516301,0.0001971718,0.00002836281,0.000003776226,0.000001401681,0.000003823034,0.0003673234],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9639512,"threshold_uncertainty_score":0.1452243,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2546609340372178,"score_gpt":0.3445541141317596,"score_spread":0.08989318009454178,"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."}}