{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:NBQXGMFI23KJVA3YUNO4ASJI5Q","short_pith_number":"pith:NBQXGMFI","schema_version":"1.0","canonical_sha256":"68617330a8d6d49a8378a35dc04928ec15e69ad1f2e2c94d8fb91cf2266b8150","source":{"kind":"arxiv","id":"2406.14541","version":3},"attestation_state":"computed","paper":{"title":"Why LLMs Are Bad at Synthetic Table Generation (and what to do about it)","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Cho-Ting Lee, Mandar Sharma, Naren Ramakrishnan, Nikhil Muralidhar, Raquib Bin Yousuf, Shengzhe Xu","submitted_at":"2024-06-20T17:52:29Z","abstract_excerpt":"Synthetic data generation is integral to ML pipelines, e.g., to augment training data, replace sensitive information, and even to power advanced platforms like DeepSeek. While LLMs fine-tuned for synthetic data generation are gaining traction, synthetic table generation -- a critical data type in business and science -- remains under-explored compared to text and image synthesis. This paper shows that LLMs, whether used as-is or after traditional fine-tuning, are inadequate for generating synthetic tables. Their autoregressive nature, combined with random order permutation during fine-tuning, "},"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":"2406.14541","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-06-20T17:52:29Z","cross_cats_sorted":[],"title_canon_sha256":"317d904acc4a8b1082334e2751725e1286a57c43ed1e993cdaacc6b4a637005b","abstract_canon_sha256":"8b54ca002b718569ce4e4996ddbb99028e777a48e0429ba02addfb2c091e0088"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:30:54.468042Z","signature_b64":"OUpquhkYJnGcOdn1aHbYyIsEX2kIO7r3BxwPQdRVP6CCcPohTyZQIV09LNhRO6gpkywrLJibcHO9EpMK3cbPAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"68617330a8d6d49a8378a35dc04928ec15e69ad1f2e2c94d8fb91cf2266b8150","last_reissued_at":"2026-07-05T10:30:54.467551Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:30:54.467551Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Why LLMs Are Bad at Synthetic Table Generation (and what to do about it)","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Cho-Ting Lee, Mandar Sharma, Naren Ramakrishnan, Nikhil Muralidhar, Raquib Bin Yousuf, Shengzhe Xu","submitted_at":"2024-06-20T17:52:29Z","abstract_excerpt":"Synthetic data generation is integral to ML pipelines, e.g., to augment training data, replace sensitive information, and even to power advanced platforms like DeepSeek. While LLMs fine-tuned for synthetic data generation are gaining traction, synthetic table generation -- a critical data type in business and science -- remains under-explored compared to text and image synthesis. This paper shows that LLMs, whether used as-is or after traditional fine-tuning, are inadequate for generating synthetic tables. Their autoregressive nature, combined with random order permutation during fine-tuning, "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.14541","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/2406.14541/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":"2406.14541","created_at":"2026-07-05T10:30:54.467611+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.14541v3","created_at":"2026-07-05T10:30:54.467611+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.14541","created_at":"2026-07-05T10:30:54.467611+00:00"},{"alias_kind":"pith_short_12","alias_value":"NBQXGMFI23KJ","created_at":"2026-07-05T10:30:54.467611+00:00"},{"alias_kind":"pith_short_16","alias_value":"NBQXGMFI23KJVA3Y","created_at":"2026-07-05T10:30:54.467611+00:00"},{"alias_kind":"pith_short_8","alias_value":"NBQXGMFI","created_at":"2026-07-05T10:30:54.467611+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.01305","citing_title":"Generative AI and Federated Learning for Intrusion Detection Systems: A Survey","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2501.01793","citing_title":"Creating Artificial Students that Never Existed: Leveraging Large Language Models and CTGANs for Synthetic Data Generation","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2503.02161","citing_title":"LLM-TabLogic: Preserving Inter-Column Logical Relationships in Synthetic Tabular Data via Prompt-Guided Latent Diffusion","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NBQXGMFI23KJVA3YUNO4ASJI5Q","json":"https://pith.science/pith/NBQXGMFI23KJVA3YUNO4ASJI5Q.json","graph_json":"https://pith.science/api/pith-number/NBQXGMFI23KJVA3YUNO4ASJI5Q/graph.json","events_json":"https://pith.science/api/pith-number/NBQXGMFI23KJVA3YUNO4ASJI5Q/events.json","paper":"https://pith.science/paper/NBQXGMFI"},"agent_actions":{"view_html":"https://pith.science/pith/NBQXGMFI23KJVA3YUNO4ASJI5Q","download_json":"https://pith.science/pith/NBQXGMFI23KJVA3YUNO4ASJI5Q.json","view_paper":"https://pith.science/paper/NBQXGMFI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.14541&json=true","fetch_graph":"https://pith.science/api/pith-number/NBQXGMFI23KJVA3YUNO4ASJI5Q/graph.json","fetch_events":"https://pith.science/api/pith-number/NBQXGMFI23KJVA3YUNO4ASJI5Q/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NBQXGMFI23KJVA3YUNO4ASJI5Q/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NBQXGMFI23KJVA3YUNO4ASJI5Q/action/storage_attestation","attest_author":"https://pith.science/pith/NBQXGMFI23KJVA3YUNO4ASJI5Q/action/author_attestation","sign_citation":"https://pith.science/pith/NBQXGMFI23KJVA3YUNO4ASJI5Q/action/citation_signature","submit_replication":"https://pith.science/pith/NBQXGMFI23KJVA3YUNO4ASJI5Q/action/replication_record"}},"created_at":"2026-07-05T10:30:54.467611+00:00","updated_at":"2026-07-05T10:30:54.467611+00:00"}