{"id":"W6929093138","doi":"10.48448/ek8y-z972","title":"On the Softmax Bottleneck of Recurrent Language Models","year":2021,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Softmax function; Perplexity; Bottleneck; Correlation; Monotonic function; Rank (graph theory); Piecewise; Matrix (chemical analysis)","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.001294912,0.0003790979,0.0004310895,0.0006692213,0.0001505196,0.00009722878,0.001692274,0.0001693303,0.004894693],"category_scores_gemma":[0.0005188035,0.000249702,0.0001200893,0.001681319,0.001627403,0.00008960725,0.0003880858,0.0004577739,0.0009060856],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002163453,"about_ca_system_score_gemma":0.0009687935,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003497844,"about_ca_topic_score_gemma":0.0006732451,"domain_scores_codex":[0.9964186,0.000123312,0.0003589423,0.0008211096,0.001706125,0.0005719188],"domain_scores_gemma":[0.997502,0.0002044554,0.0004741597,0.001493644,0.000183764,0.0001420175],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002647625,0.0006504644,0.000009744813,0.0001591096,0.00009289369,0.00003850935,0.001265401,0.001607344,0.01382019,0.2069644,0.7610624,0.01430307],"study_design_scores_gemma":[0.003580592,0.001684352,0.0001134844,0.01333116,0.0005320972,0.0001129464,0.0088485,0.4061826,0.03496732,0.06784043,0.4567067,0.006099772],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.001465924,0.003677618,0.001331382,0.0004174514,0.0009671996,0.0006899388,0.0005346204,0.0002891891,0.9906267],"genre_scores_gemma":[0.2408206,0.0003682507,0.01316627,0.001278739,0.001093799,0.0001060771,0.0004037102,0.002300974,0.7404616],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.4045753,"threshold_uncertainty_score":0.9999955,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04317076503652927,"score_gpt":0.3134520370440302,"score_spread":0.2702812720075009,"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."}}