{"id":"W4416035393","doi":"10.18653/v1/2025.emnlp-main.1506","title":"Machine-generated text detection prevents language model collapse","year":2025,"lang":"en","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Engineering and Physical Sciences Research Council","keywords":"Language model; Natural language; Language identification; Noise (video); Field (mathematics)","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":[0.006059159,0.001847034,0.001429479,0.00143859,0.0009419718,0.001545544,0.001983605,0.002140942,0.001144157],"category_scores_gemma":[0.03934618,0.0006597677,0.001047568,0.0006772174,0.001429317,0.002970074,0.002421868,0.002714008,0.001716816],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001009905,"about_ca_system_score_gemma":0.001997855,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004334304,"about_ca_topic_score_gemma":0.005318798,"domain_scores_codex":[0.9970298,0.001167404,0.0001744539,0.000868932,0.0005378364,0.0002215649],"domain_scores_gemma":[0.9806486,0.01224226,0.001047463,0.0032349,0.00235831,0.0004685529],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001329684,0.0008209496,0.02103554,0.0005857312,0.0003937645,0.0007593759,0.001132279,0.4170237,0.06245198,0.00526444,0.01296384,0.4762387],"study_design_scores_gemma":[0.00003673976,0.0001893998,0.001325765,0.00002771681,0.00003184766,0.0001808486,0.00007713924,0.9594626,0.03514546,0.001941173,0.001553754,0.00002750117],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3601703,0.001020002,0.6168424,0.0008449104,0.0002851067,0.0002837437,0.0006008974,0.01741112,0.002541429],"genre_scores_gemma":[0.8134007,0.0001733207,0.1797614,0.0005425174,0.0001087138,0.0002106286,0.00249186,0.0009808157,0.002330065],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.006059159,"threshold_uncertainty_score":0.03204423,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0121339640939598,"score_gpt":0.2561407342494015,"score_spread":0.2440067701554417,"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."}}