{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:A67KMUY5PYDJGUIKI3EDESC6C5","short_pith_number":"pith:A67KMUY5","schema_version":"1.0","canonical_sha256":"07bea6531d7e0693510a46c832485e1773aed0f9df63e00248a07af2c07741bb","source":{"kind":"arxiv","id":"2505.20161","version":1},"attestation_state":"computed","paper":{"title":"Prismatic Synthesis: Gradient-based Data Diversification Boosts Generalization in LLM Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Bryan Catanzaro, David Acuna, Jaehun Jung, Mohammad Shoeybi, Mostafa Patwary, Seungju Han, Shrimai Prabhumoye, Skyler Hallinan, Ximing Lu, Yejin Choi","submitted_at":"2025-05-26T16:05:10Z","abstract_excerpt":"Effective generalization in language models depends critically on the diversity of their training data. Yet existing diversity metrics often fall short of this goal, relying on surface-level heuristics that are decoupled from model behavior. This motivates us to ask: What kind of diversity in training data actually drives generalization in language models -- and how can we measure and amplify it? Through large-scale empirical analyses spanning over 300 training runs, carefully controlled for data scale and quality, we show that data diversity can be a strong predictor of generalization in LLM "},"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":"2505.20161","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-26T16:05:10Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"8f4fd79196329fded6f75be299f48a44a741718c90a6e4154e8e9e8e193c1c6a","abstract_canon_sha256":"0d629756878a422d0cc689d1186cd990e0b08ae32e475fc4a9c3ec5969a03a6f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:09:53.551729Z","signature_b64":"K6nuyHM5UQQcwbL3Ji5iwgiin7witnQ8wTK9VeX+kVBIaZcbVA7oF6L407Dd2FRPxgP/BRpD/9b1ePyt5hc8Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"07bea6531d7e0693510a46c832485e1773aed0f9df63e00248a07af2c07741bb","last_reissued_at":"2026-07-05T11:09:53.551231Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:09:53.551231Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Prismatic Synthesis: Gradient-based Data Diversification Boosts Generalization in LLM Reasoning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Bryan Catanzaro, David Acuna, Jaehun Jung, Mohammad Shoeybi, Mostafa Patwary, Seungju Han, Shrimai Prabhumoye, Skyler Hallinan, Ximing Lu, Yejin Choi","submitted_at":"2025-05-26T16:05:10Z","abstract_excerpt":"Effective generalization in language models depends critically on the diversity of their training data. Yet existing diversity metrics often fall short of this goal, relying on surface-level heuristics that are decoupled from model behavior. This motivates us to ask: What kind of diversity in training data actually drives generalization in language models -- and how can we measure and amplify it? Through large-scale empirical analyses spanning over 300 training runs, carefully controlled for data scale and quality, we show that data diversity can be a strong predictor of generalization in LLM "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.20161","kind":"arxiv","version":1},"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/2505.20161/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":"2505.20161","created_at":"2026-07-05T11:09:53.551286+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.20161v1","created_at":"2026-07-05T11:09:53.551286+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.20161","created_at":"2026-07-05T11:09:53.551286+00:00"},{"alias_kind":"pith_short_12","alias_value":"A67KMUY5PYDJ","created_at":"2026-07-05T11:09:53.551286+00:00"},{"alias_kind":"pith_short_16","alias_value":"A67KMUY5PYDJGUIK","created_at":"2026-07-05T11:09:53.551286+00:00"},{"alias_kind":"pith_short_8","alias_value":"A67KMUY5","created_at":"2026-07-05T11:09:53.551286+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":5,"sample":[{"citing_arxiv_id":"2607.02423","citing_title":"Neuron-Aware Active Few-Shot Learning for LLMs","ref_index":46,"is_internal_anchor":true},{"citing_arxiv_id":"2606.12385","citing_title":"Which Models Are Our Models Built On? Auditing Invisible Dependencies in Modern LLMs","ref_index":30,"is_internal_anchor":true},{"citing_arxiv_id":"2605.29448","citing_title":"How Much Is a Dataset Worth? Scaling Laws, the Vendi Score, and Matrix Spectral Functions","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2604.08519","citing_title":"Cram Less to Fit More: Training Data Pruning Improves Memorization of Facts","ref_index":40,"is_internal_anchor":true},{"citing_arxiv_id":"2604.14564","citing_title":"MARS$^2$: Scaling Multi-Agent Tree Search via Reinforcement Learning for Code Generation","ref_index":10,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/A67KMUY5PYDJGUIKI3EDESC6C5","json":"https://pith.science/pith/A67KMUY5PYDJGUIKI3EDESC6C5.json","graph_json":"https://pith.science/api/pith-number/A67KMUY5PYDJGUIKI3EDESC6C5/graph.json","events_json":"https://pith.science/api/pith-number/A67KMUY5PYDJGUIKI3EDESC6C5/events.json","paper":"https://pith.science/paper/A67KMUY5"},"agent_actions":{"view_html":"https://pith.science/pith/A67KMUY5PYDJGUIKI3EDESC6C5","download_json":"https://pith.science/pith/A67KMUY5PYDJGUIKI3EDESC6C5.json","view_paper":"https://pith.science/paper/A67KMUY5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.20161&json=true","fetch_graph":"https://pith.science/api/pith-number/A67KMUY5PYDJGUIKI3EDESC6C5/graph.json","fetch_events":"https://pith.science/api/pith-number/A67KMUY5PYDJGUIKI3EDESC6C5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/A67KMUY5PYDJGUIKI3EDESC6C5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/A67KMUY5PYDJGUIKI3EDESC6C5/action/storage_attestation","attest_author":"https://pith.science/pith/A67KMUY5PYDJGUIKI3EDESC6C5/action/author_attestation","sign_citation":"https://pith.science/pith/A67KMUY5PYDJGUIKI3EDESC6C5/action/citation_signature","submit_replication":"https://pith.science/pith/A67KMUY5PYDJGUIKI3EDESC6C5/action/replication_record"}},"created_at":"2026-07-05T11:09:53.551286+00:00","updated_at":"2026-07-05T11:09:53.551286+00:00"}