{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:VMFX7YHVPVDO7H7XUQUBKXHHKJ","short_pith_number":"pith:VMFX7YHV","schema_version":"1.0","canonical_sha256":"ab0b7fe0f57d46ef9ff7a428155ce75268787541e126c5bd8bc6742e02b8e062","source":{"kind":"arxiv","id":"2412.14689","version":3},"attestation_state":"computed","paper":{"title":"How to Synthesize Text Data without Model Collapse?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bowen Zhou, Daixuan Cheng, Ermo Hua, Hengli Li, Kaiyan Zhang, Ning Ding, Xingtai Lv, Xuekai Zhu, Zhouhan Lin, Zilong Zheng","submitted_at":"2024-12-19T09:43:39Z","abstract_excerpt":"Model collapse in synthetic data indicates that iterative training on self-generated data leads to a gradual decline in performance. With the proliferation of AI models, synthetic data will fundamentally reshape the web data ecosystem. Future GPT-$\\{n\\}$ models will inevitably be trained on a blend of synthetic and human-produced data. In this paper, we focus on two questions: what is the impact of synthetic data on language model training, and how to synthesize data without model collapse? We first pre-train language models across different proportions of synthetic data, revealing a negative "},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2412.14689","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-19T09:43:39Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"73ec5819b38d42cfe5fdf4a9e6abd6eda5d2251a93ed6d666bf1dac570984b9a","abstract_canon_sha256":"e91d91c0d6d8382420e94c820b98f689df9d007060b8f232ba5d9b7bf18913e8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:49.684745Z","signature_b64":"NGV9T8ftijDctcdjz6xrL2QhqgQJJ8unR09urw58c3oKFeMP0N5Md2HPhUP4JeKKMMQktqNQFl5gzYJ/FbG/BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ab0b7fe0f57d46ef9ff7a428155ce75268787541e126c5bd8bc6742e02b8e062","last_reissued_at":"2026-07-05T11:10:49.684185Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:49.684185Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How to Synthesize Text Data without Model Collapse?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Bowen Zhou, Daixuan Cheng, Ermo Hua, Hengli Li, Kaiyan Zhang, Ning Ding, Xingtai Lv, Xuekai Zhu, Zhouhan Lin, Zilong Zheng","submitted_at":"2024-12-19T09:43:39Z","abstract_excerpt":"Model collapse in synthetic data indicates that iterative training on self-generated data leads to a gradual decline in performance. With the proliferation of AI models, synthetic data will fundamentally reshape the web data ecosystem. Future GPT-$\\{n\\}$ models will inevitably be trained on a blend of synthetic and human-produced data. In this paper, we focus on two questions: what is the impact of synthetic data on language model training, and how to synthesize data without model collapse? We first pre-train language models across different proportions of synthetic data, revealing a negative "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.14689","kind":"arxiv","version":3},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2412.14689/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2412.14689","created_at":"2026-07-05T11:10:49.684253+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.14689v3","created_at":"2026-07-05T11:10:49.684253+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.14689","created_at":"2026-07-05T11:10:49.684253+00:00"},{"alias_kind":"pith_short_12","alias_value":"VMFX7YHVPVDO","created_at":"2026-07-05T11:10:49.684253+00:00"},{"alias_kind":"pith_short_16","alias_value":"VMFX7YHVPVDO7H7X","created_at":"2026-07-05T11:10:49.684253+00:00"},{"alias_kind":"pith_short_8","alias_value":"VMFX7YHV","created_at":"2026-07-05T11:10:49.684253+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.28438","citing_title":"When AI Reviews Its Own Code: Recursive Self-Training Collapse in Code LLMs","ref_index":156,"is_internal_anchor":false},{"citing_arxiv_id":"2502.04419","citing_title":"Understanding and Mitigating Bias Inheritance in LLM-based Data Augmentation on Downstream Tasks","ref_index":40,"is_internal_anchor":false},{"citing_arxiv_id":"2503.17181","citing_title":"A Study of LLMs' Preferences for Libraries and Programming Languages","ref_index":87,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04127","citing_title":"Position: the Stochastic Parrot in the Coal Mine. Model Collapse is a Threat to Low-Resource Communities","ref_index":37,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VMFX7YHVPVDO7H7XUQUBKXHHKJ","json":"https://pith.science/pith/VMFX7YHVPVDO7H7XUQUBKXHHKJ.json","graph_json":"https://pith.science/api/pith-number/VMFX7YHVPVDO7H7XUQUBKXHHKJ/graph.json","events_json":"https://pith.science/api/pith-number/VMFX7YHVPVDO7H7XUQUBKXHHKJ/events.json","paper":"https://pith.science/paper/VMFX7YHV"},"agent_actions":{"view_html":"https://pith.science/pith/VMFX7YHVPVDO7H7XUQUBKXHHKJ","download_json":"https://pith.science/pith/VMFX7YHVPVDO7H7XUQUBKXHHKJ.json","view_paper":"https://pith.science/paper/VMFX7YHV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.14689&json=true","fetch_graph":"https://pith.science/api/pith-number/VMFX7YHVPVDO7H7XUQUBKXHHKJ/graph.json","fetch_events":"https://pith.science/api/pith-number/VMFX7YHVPVDO7H7XUQUBKXHHKJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VMFX7YHVPVDO7H7XUQUBKXHHKJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VMFX7YHVPVDO7H7XUQUBKXHHKJ/action/storage_attestation","attest_author":"https://pith.science/pith/VMFX7YHVPVDO7H7XUQUBKXHHKJ/action/author_attestation","sign_citation":"https://pith.science/pith/VMFX7YHVPVDO7H7XUQUBKXHHKJ/action/citation_signature","submit_replication":"https://pith.science/pith/VMFX7YHVPVDO7H7XUQUBKXHHKJ/action/replication_record"}},"created_at":"2026-07-05T11:10:49.684253+00:00","updated_at":"2026-07-05T11:10:49.684253+00:00"}