{"id":"W6966746228","doi":"10.48448/pewt-ca73","title":"On the Utility of Pretraining Language Models on Synthetic Data","year":2024,"lang":"en","type":"other","venue":"Underline Science Inc.","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"","keywords":"Synthetic data; Language model; Leverage (statistics); Training set; Data modeling; Machine translation; Generative model","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":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.004020025,0.0003888487,0.000380511,0.0007830306,0.0001113432,0.000160447,0.004159014,0.0001697341,0.003966092],"category_scores_gemma":[0.001328886,0.0002420696,0.00006167808,0.001387784,0.002476279,0.000196042,0.0009838125,0.0006915627,0.004691011],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000101599,"about_ca_system_score_gemma":0.000563684,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003225113,"about_ca_topic_score_gemma":0.0002999635,"domain_scores_codex":[0.9957588,0.0001423963,0.0003534376,0.001453266,0.001752113,0.0005399921],"domain_scores_gemma":[0.9946211,0.00037978,0.0003010238,0.00451782,0.00006479015,0.0001154798],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00003381541,0.0003022262,0.000001482902,0.0002147938,0.00008725737,0.00003248081,0.001865892,0.0002576432,0.001166947,0.08534941,0.897999,0.01268901],"study_design_scores_gemma":[0.0001303598,0.0001285996,0.000003462009,0.001945278,0.00009156406,0.000007495845,0.001009243,0.9700421,0.0004501213,0.01398045,0.01179658,0.0004147142],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.0003994208,0.0005156989,0.0006932876,0.0002485789,0.0005414675,0.0005951285,0.003432151,0.0004256162,0.9931486],"genre_scores_gemma":[0.8948375,0.00001086183,0.001490957,0.0002258884,0.0001801687,0.00001259222,0.0001236986,0.0007202472,0.1023981],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.9697845,"threshold_uncertainty_score":0.9969444,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1269172173821471,"score_gpt":0.3596625402523217,"score_spread":0.2327453228701746,"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."}}