{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003588298,0.002203419,0.0006449921,0.0008493225,0.0006329245,0.001366434,0.00157954,0.001592603,0.003819672],"category_scores_gemma":[0.01630306,0.0006293718,0.0007629487,0.0008162825,0.0008855581,0.00306829,0.001499526,0.002198156,0.002626496],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000892878,"about_ca_system_score_gemma":0.001235149,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01493225,"about_ca_topic_score_gemma":0.02272561,"domain_scores_codex":[0.998078,0.001107659,0.00009879834,0.0004144041,0.0002063107,0.0000948826],"domain_scores_gemma":[0.9925371,0.004967138,0.0002018825,0.001276227,0.0008839815,0.0001336759],"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":[0.0006256185,0.0006339688,0.009079032,0.0005243755,0.00044043,0.0002870945,0.0002055827,0.723669,0.01630116,0.003079188,0.01569673,0.2294579],"study_design_scores_gemma":[0.00005377665,0.0002782861,0.002244062,0.00007752424,0.00005415187,0.0001058425,0.000134709,0.9691373,0.0204606,0.002257806,0.005148386,0.00004763775],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6293158,0.004280724,0.3180492,0.002695039,0.0007491955,0.0004867814,0.008775001,0.01572781,0.01992035],"genre_scores_gemma":[0.8426135,0.00114295,0.1237923,0.0007733725,0.0001015879,0.0003995769,0.02579153,0.0006439847,0.004741267],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01493225,"threshold_uncertainty_score":0.02969068,"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."}}