{"id":"W4393968110","doi":"10.48550/arxiv.2404.02305","title":"Collapse of Self-trained Language Models","year":2024,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Institute for Catastrophic Loss Reduction","keywords":"Linguistics; Computer science; Natural language processing; Psychology; Philosophy","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004010954,0.00144882,0.001462837,0.0008951404,0.0006905507,0.001817893,0.002622025,0.00193894,0.003884476],"category_scores_gemma":[0.02641086,0.001355114,0.001406841,0.0007440389,0.00174226,0.006391633,0.005059617,0.00512263,0.002469971],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001339883,"about_ca_system_score_gemma":0.00161542,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0055786,"about_ca_topic_score_gemma":0.00722689,"domain_scores_codex":[0.9977227,0.0008646669,0.000105081,0.0007031603,0.0004141926,0.0001902643],"domain_scores_gemma":[0.9874493,0.006970429,0.0004198895,0.003416773,0.001382005,0.0003616416],"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":[0.0005504463,0.0002644083,0.005275075,0.0002246321,0.0002709034,0.0004088051,0.0007615994,0.7656358,0.01418417,0.01353509,0.006394311,0.1924947],"study_design_scores_gemma":[0.000009000137,0.00004873062,0.0001719589,0.00001091497,0.00001321815,0.00004068254,0.0000250597,0.9874359,0.003814412,0.007927205,0.0004921067,0.00001076901],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1354709,0.0004833419,0.8488787,0.0008568849,0.0001908349,0.0001407354,0.0003926929,0.0102524,0.003333445],"genre_scores_gemma":[0.83069,0.0002007336,0.1593148,0.0007063281,0.00008547944,0.0002141079,0.00169578,0.001272905,0.00581984],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0055786,"threshold_uncertainty_score":0.02121222,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03686683997944842,"score_gpt":0.2020561457262378,"score_spread":0.1651893057467894,"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."}}