{"id":"W4385571741","doi":"10.18653/v1/2023.acl-short.74","title":"Gradient Ascent Post-training Enhances Language Model Generalization","year":2023,"lang":"en","type":"article","venue":"","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Kootenay Association for Science & Technology","funders":"Institute for Information and Communications Technology Promotion; Ministry of Science and ICT, South Korea; Korea Advanced Institute of Science and Technology","keywords":"Generalization; Computer science; Task (project management); Artificial intelligence; Training set; Language model; Training (meteorology); Zero (linguistics); Natural language processing; Speech recognition; Mathematics; Linguistics","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":[0.002023574,0.001632194,0.0008317164,0.0005910173,0.000444556,0.0008687178,0.00123972,0.001417888,0.004409952],"category_scores_gemma":[0.01061663,0.0004284755,0.0005807791,0.0004121774,0.0006545887,0.002208801,0.001392544,0.002687757,0.003059825],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005923008,"about_ca_system_score_gemma":0.0009422289,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004605994,"about_ca_topic_score_gemma":0.008230888,"domain_scores_codex":[0.9990295,0.0002591159,0.00004944139,0.0003716809,0.0001827258,0.0001075823],"domain_scores_gemma":[0.9976184,0.001306499,0.0000808332,0.000475287,0.0004097668,0.0001090971],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0005228193,0.0005766494,0.003275865,0.0003143125,0.0001982908,0.0002807852,0.0003172563,0.2693488,0.07269815,0.002331396,0.0150569,0.6350787],"study_design_scores_gemma":[0.00003076492,0.0001856826,0.002069536,0.00001888768,0.00003547463,0.00009461219,0.00005697936,0.9652613,0.02744902,0.001926304,0.002846615,0.00002482478],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2534217,0.001877186,0.7072279,0.001013494,0.000645932,0.0002040571,0.0005571331,0.02573957,0.009313061],"genre_scores_gemma":[0.7990459,0.00036571,0.1882223,0.000680727,0.0001556604,0.0001716191,0.001691919,0.001107116,0.008559028],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.004605994,"threshold_uncertainty_score":0.01475269,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02295383156216873,"score_gpt":0.2952370719509481,"score_spread":0.2722832403887794,"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."}}