{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01058386,0.0007562826,0.001140432,0.002098244,0.001135178,0.006208654,0.001581471,0.001482432,0.002405102],"category_scores_gemma":[0.07021758,0.0008560536,0.001182214,0.001007939,0.005113299,0.01792755,0.005546941,0.003525001,0.0003077607],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001437486,"about_ca_system_score_gemma":0.0007369459,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001606131,"about_ca_topic_score_gemma":0.001143675,"domain_scores_codex":[0.9950294,0.003049746,0.000271349,0.0007310906,0.0006984283,0.0002200267],"domain_scores_gemma":[0.9548239,0.03236895,0.00342927,0.006920514,0.001433057,0.0010243],"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":[0.000700598,0.0002233741,0.03978578,0.0004011853,0.0004294364,0.0003709832,0.01263272,0.1851702,0.01934673,0.5971087,0.001137059,0.1426932],"study_design_scores_gemma":[0.00002234669,0.0001366105,0.003793565,0.00004159658,0.00004442297,0.0002048481,0.0009201876,0.5601246,0.003417427,0.4297125,0.001519022,0.00006280609],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2829188,0.0003682679,0.7093296,0.001497902,0.0000348909,0.00006277303,0.0001336704,0.0003418945,0.005312146],"genre_scores_gemma":[0.9284614,0.0001487133,0.07052887,0.0000827289,0.00002713317,0.0000609038,0.0001251845,0.0001447132,0.000420336],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01058386,"threshold_uncertainty_score":0.05597347,"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."}}