{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002418596,0.000629329,0.0007546476,0.0009304287,0.0006246113,0.003120348,0.001371996,0.0008619099,0.0050507],"category_scores_gemma":[0.01936397,0.0007772353,0.001254379,0.0008313852,0.002256432,0.00822147,0.002552955,0.002746642,0.00209832],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00180369,"about_ca_system_score_gemma":0.001293432,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00517139,"about_ca_topic_score_gemma":0.004993743,"domain_scores_codex":[0.9972573,0.001190216,0.0001580894,0.0007423502,0.0004130982,0.0002389272],"domain_scores_gemma":[0.9900362,0.007323484,0.0004521124,0.001499172,0.0005225147,0.0001665059],"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":[0.0002386028,0.00007350083,0.002128589,0.0002025057,0.0000724774,0.0002912892,0.00222915,0.1198743,0.006633856,0.7784848,0.003089566,0.08668137],"study_design_scores_gemma":[0.00001384005,0.00003184044,0.0002216446,0.00002070463,0.00002243483,0.00008312349,0.0001315546,0.3859794,0.00266043,0.6094799,0.001331489,0.00002363145],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05228995,0.0001542135,0.9373389,0.0006021623,0.00002590017,0.00004725887,0.0003655188,0.0013711,0.007804973],"genre_scores_gemma":[0.8573452,0.0002896195,0.1330137,0.0004105691,0.00007872289,0.0001719112,0.001252511,0.0007820965,0.006655677],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.00517139,"threshold_uncertainty_score":0.01689625,"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."}}