{"id":"W4400949264","doi":"10.1038/s41586-024-07566-y","title":"AI models collapse when trained on recursively generated data","year":2024,"lang":"en","type":"article","venue":"Nature","topic":"Topic Modeling","field":"Computer Science","cited_by":583,"is_retracted":false,"has_abstract":true,"ca_institutions":"Vector Institute; University of Toronto","funders":"","keywords":"Generative grammar; Generative model; Intuition; Computer science; Variety (cybernetics); The Internet; Artificial intelligence; Data science; Psychology; Cognitive science; World Wide Web","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002722175,0.0008498321,0.0006725438,0.0007067595,0.0004652006,0.00128953,0.001320143,0.001493062,0.003209789],"category_scores_gemma":[0.01098684,0.0006960903,0.001163478,0.0006535237,0.0007109308,0.001979758,0.001246835,0.003530083,0.002546047],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001491201,"about_ca_system_score_gemma":0.001264034,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01773535,"about_ca_topic_score_gemma":0.02312865,"domain_scores_codex":[0.9992014,0.0002949761,0.00004611061,0.000282748,0.0001086898,0.00006620877],"domain_scores_gemma":[0.996143,0.002368566,0.0001219655,0.0008072818,0.0004578155,0.0001013999],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.000181842,0.0001825316,0.007565845,0.0001867833,0.0003035087,0.00012224,0.0003487478,0.7034908,0.004747043,0.01621577,0.01024293,0.2564119],"study_design_scores_gemma":[0.000005852139,0.00002878201,0.0005506502,0.0000123686,0.0000116936,0.00002292409,0.00001828031,0.9916589,0.001391678,0.004964153,0.001327402,0.000007424414],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2919036,0.003011813,0.668986,0.002879228,0.000740539,0.0002357117,0.001571375,0.01091777,0.01975393],"genre_scores_gemma":[0.8662288,0.0005348307,0.119042,0.0007439231,0.0001159214,0.0001681255,0.002561962,0.0005013198,0.01010317],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9972778,"threshold_uncertainty_score":0.03526425,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0509041440660831,"score_gpt":0.3005333002021868,"score_spread":0.2496291561361037,"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."}}