{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CGWZ3LAXF7HEEVZTVUME2VJX5W","short_pith_number":"pith:CGWZ3LAX","schema_version":"1.0","canonical_sha256":"11ad9dac172fce425733ad184d5537ed888992d656831538d49cc92286ed4ef1","source":{"kind":"arxiv","id":"2507.20181","version":1},"attestation_state":"computed","paper":{"title":"SGPO: Self-Generated Preference Optimization based on Self-Improver","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Daejin Jo, Hyeonji Lee, Seohwan Yun, Sungwoong Kim","submitted_at":"2025-07-27T08:55:40Z","abstract_excerpt":"Large language models (LLMs), despite their extensive pretraining on diverse datasets, require effective alignment to human preferences for practical and reliable deployment. Conventional alignment methods typically employ off-policy learning and depend on human-annotated datasets, which limits their broad applicability and introduces distribution shift issues during training. To address these challenges, we propose Self-Generated Preference Optimization based on Self-Improver (SGPO), an innovative alignment framework that leverages an on-policy self-improving mechanism. Specifically, the impr"},"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":"2507.20181","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-27T08:55:40Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cc5edb0fb812fb525ddf612a7cf5ff65908fd7c725a8d8fb771a33764adfe385","abstract_canon_sha256":"aa46464b21753ea94535f909c9c75cd8c545d09cf64031b7e53392385655dc74"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:44:14.783566Z","signature_b64":"P4CwJUQ1Q9oIUQYc/2Prs2VULTFDcJotHED2yOibJvBT2qrny3oWgFrOJn4J0/Z8UPmnBLm7M9YOvD0EKLCQDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"11ad9dac172fce425733ad184d5537ed888992d656831538d49cc92286ed4ef1","last_reissued_at":"2026-07-05T11:44:14.783092Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:44:14.783092Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SGPO: Self-Generated Preference Optimization based on Self-Improver","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Daejin Jo, Hyeonji Lee, Seohwan Yun, Sungwoong Kim","submitted_at":"2025-07-27T08:55:40Z","abstract_excerpt":"Large language models (LLMs), despite their extensive pretraining on diverse datasets, require effective alignment to human preferences for practical and reliable deployment. Conventional alignment methods typically employ off-policy learning and depend on human-annotated datasets, which limits their broad applicability and introduces distribution shift issues during training. To address these challenges, we propose Self-Generated Preference Optimization based on Self-Improver (SGPO), an innovative alignment framework that leverages an on-policy self-improving mechanism. Specifically, the impr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.20181","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/2507.20181/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":"2507.20181","created_at":"2026-07-05T11:44:14.783154+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.20181v1","created_at":"2026-07-05T11:44:14.783154+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.20181","created_at":"2026-07-05T11:44:14.783154+00:00"},{"alias_kind":"pith_short_12","alias_value":"CGWZ3LAXF7HE","created_at":"2026-07-05T11:44:14.783154+00:00"},{"alias_kind":"pith_short_16","alias_value":"CGWZ3LAXF7HEEVZT","created_at":"2026-07-05T11:44:14.783154+00:00"},{"alias_kind":"pith_short_8","alias_value":"CGWZ3LAX","created_at":"2026-07-05T11:44:14.783154+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.25511","citing_title":"CRPO: Character-centric Group Relative Policy Optimization for Role-aware Reasoning in Role-playing Agents","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CGWZ3LAXF7HEEVZTVUME2VJX5W","json":"https://pith.science/pith/CGWZ3LAXF7HEEVZTVUME2VJX5W.json","graph_json":"https://pith.science/api/pith-number/CGWZ3LAXF7HEEVZTVUME2VJX5W/graph.json","events_json":"https://pith.science/api/pith-number/CGWZ3LAXF7HEEVZTVUME2VJX5W/events.json","paper":"https://pith.science/paper/CGWZ3LAX"},"agent_actions":{"view_html":"https://pith.science/pith/CGWZ3LAXF7HEEVZTVUME2VJX5W","download_json":"https://pith.science/pith/CGWZ3LAXF7HEEVZTVUME2VJX5W.json","view_paper":"https://pith.science/paper/CGWZ3LAX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.20181&json=true","fetch_graph":"https://pith.science/api/pith-number/CGWZ3LAXF7HEEVZTVUME2VJX5W/graph.json","fetch_events":"https://pith.science/api/pith-number/CGWZ3LAXF7HEEVZTVUME2VJX5W/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CGWZ3LAXF7HEEVZTVUME2VJX5W/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CGWZ3LAXF7HEEVZTVUME2VJX5W/action/storage_attestation","attest_author":"https://pith.science/pith/CGWZ3LAXF7HEEVZTVUME2VJX5W/action/author_attestation","sign_citation":"https://pith.science/pith/CGWZ3LAXF7HEEVZTVUME2VJX5W/action/citation_signature","submit_replication":"https://pith.science/pith/CGWZ3LAXF7HEEVZTVUME2VJX5W/action/replication_record"}},"created_at":"2026-07-05T11:44:14.783154+00:00","updated_at":"2026-07-05T11:44:14.783154+00:00"}