{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002357904,0.0001272654,0.0001121216,0.0001010458,0.00006243731,0.0003086839,0.001636961,0.0005257963,0.00001755915],"category_scores_gemma":[0.00004397265,0.0001068325,0.00003040709,0.0003458491,0.00001136389,0.0006887991,0.000318677,0.001417031,0.00004446961],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004933295,"about_ca_system_score_gemma":0.0001718681,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008952771,"about_ca_topic_score_gemma":0.000009625984,"domain_scores_codex":[0.9986193,0.00004292385,0.0001303119,0.0006778495,0.0003200958,0.000209478],"domain_scores_gemma":[0.9986145,0.00005101007,0.00001870601,0.001173451,0.00007210367,0.00007020986],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001167696,0.00003346723,0.000001048428,0.00003220093,0.0000453824,0.0001728294,0.001469124,0.008399446,0.0009933041,0.5033867,0.4207263,0.06472844],"study_design_scores_gemma":[0.000115549,0.00002187937,0.000003312738,0.00005338455,0.000005313222,0.00001052318,0.000004660924,0.9038107,0.0003648047,0.03730826,0.05817962,0.0001219896],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.003036857,0.004796182,0.931289,0.05128295,0.00287191,0.0002809689,0.0001191978,0.000746467,0.005576429],"genre_scores_gemma":[0.8798167,0.00003946312,0.1006428,0.01532151,0.0006333007,0.000009513549,0.0001034666,0.00002751944,0.003405756],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.8954113,"threshold_uncertainty_score":0.6156372,"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."}}