{"id":"W4386043297","doi":"10.48550/arxiv.2308.09124","title":"Linearity of Relation Decoding in Transformer Language Models","year":2023,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Topic Modeling","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Azrieli Foundation; Open Philanthropy Project","keywords":"Transformer; Linearity; Computer science; Decoding methods; Representation (politics); Computation; Relation (database); Variety (cybernetics); Artificial intelligence; Natural language processing; Theoretical computer science; Algorithm; Data mining; Electronic engineering","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":[],"consensus_categories":[],"category_scores_codex":[0.0003375029,0.0001402555,0.000228973,0.000332123,0.0000324147,0.00002458442,0.0008252594,0.000221252,0.0000047305],"category_scores_gemma":[0.00002094221,0.0001743023,0.0001134543,0.0004614622,0.00002511834,0.0004004542,0.0004005089,0.0004332922,0.00001250474],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001249233,"about_ca_system_score_gemma":0.0001044856,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008535773,"about_ca_topic_score_gemma":0.0003528486,"domain_scores_codex":[0.9988351,0.00006663833,0.0002285463,0.000594986,0.00007863205,0.0001961046],"domain_scores_gemma":[0.9990367,0.0000758321,0.0001259197,0.000659699,0.00005167324,0.0000501237],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000005461191,0.00002055772,0.002089668,0.00006447281,0.00001217171,0.00006207829,0.001494122,0.890962,0.00005980602,0.1044457,0.000003707268,0.000780319],"study_design_scores_gemma":[0.0002158582,0.000006968949,0.001013325,0.0001046927,0.00001034923,4.626398e-7,0.000084343,0.9273336,0.0001756333,0.0709064,0.000002773532,0.0001455715],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3958612,0.00001961944,0.6029013,0.00003593667,0.0001753623,0.0001065063,0.000003581041,0.00009087344,0.00080566],"genre_scores_gemma":[0.9920998,0.00007620216,0.007293632,0.00001213066,0.00002173082,4.568028e-7,0.000006872921,0.00001024614,0.0004789509],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5962386,"threshold_uncertainty_score":0.7107843,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.153890287573156,"score_gpt":0.2144195023585443,"score_spread":0.06052921478538839,"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."}}