{"id":"W4400702583","doi":"10.52202/079017-3256","title":"Self-Consuming Generative Models with Curated Data Provably Optimize Human Preferences","year":2024,"lang":"en","type":"article","venue":"","topic":"Generative Adversarial Networks and Image Synthesis","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Canadian Institute for Advanced Research; Université de Montréal; Mila - Quebec Artificial Intelligence Institute","funders":"","keywords":"Generative model; Computer science; Generative grammar; Retraining; Pairwise comparison; Upload; Artificial intelligence; Function (biology); Machine learning; Stability (learning theory); Synthetic data","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":[0.004748957,0.001534349,0.001466773,0.0007209438,0.0007998645,0.00194194,0.002603802,0.002249619,0.003861734],"category_scores_gemma":[0.03149091,0.001225391,0.001344903,0.0006071256,0.00238872,0.00314272,0.002997019,0.003414224,0.0009969914],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001715612,"about_ca_system_score_gemma":0.00124111,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003845524,"about_ca_topic_score_gemma":0.004644444,"domain_scores_codex":[0.998082,0.0008603794,0.00007660211,0.0004987422,0.0002922568,0.0001900415],"domain_scores_gemma":[0.9798982,0.0159298,0.0009040748,0.001952535,0.0007818059,0.0005335283],"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":[0.0002111597,0.0001165286,0.002425478,0.0001573278,0.00009311166,0.000245783,0.0004584526,0.8582907,0.003920563,0.09451769,0.002992746,0.03657054],"study_design_scores_gemma":[0.00001375811,0.00002881216,0.0001186011,0.00001181843,0.00001007456,0.0000430863,0.00001796663,0.9668678,0.0006733504,0.03169338,0.0005108592,0.00001045393],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.04404318,0.0003308615,0.9510692,0.0005944913,0.00003537081,0.00009413855,0.00009607339,0.0006615548,0.003075171],"genre_scores_gemma":[0.7809686,0.0004085858,0.2056156,0.0006961546,0.0001041962,0.0004728258,0.0004161371,0.0007232597,0.01059466],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.004748957,"threshold_uncertainty_score":0.02511519,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06929622210286662,"score_gpt":0.2768607772941121,"score_spread":0.2075645551912455,"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."}}