{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:OGZPKQGTCSENJWJPZC43VTHTWU","short_pith_number":"pith:OGZPKQGT","schema_version":"1.0","canonical_sha256":"71b2f540d31488d4d92fc8b9baccf3b50b1e0c4e3b7fa55b410c8ec2c178c176","source":{"kind":"arxiv","id":"2312.10584","version":2},"attestation_state":"computed","paper":{"title":"Policy Optimization in RLHF: The Impact of Out-of-preference Data","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Tian Xu, Yang Yu, Ziniu Li","submitted_at":"2023-12-17T02:14:15Z","abstract_excerpt":"Aligning intelligent agents with human preferences and values is important. This paper examines two popular alignment methods: Direct Preference Optimization (DPO) and Reward-Model-Based Policy Optimization (RMB-PO). A variant of RMB-PO, referred to as RMB-PO+ is also considered. These methods, either explicitly or implicitly, learn a reward model from preference data and differ in the data used for policy optimization to unlock the generalization ability of the reward model. In particular, compared with DPO, RMB-PO additionally uses policy-generated data, and RMB-PO+ further leverages new, pr"},"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":"2312.10584","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2023-12-17T02:14:15Z","cross_cats_sorted":[],"title_canon_sha256":"584b9ac1112af26c04fb614f7cf26450b2070c5d7f0f2b78ffc4ce96061d3204","abstract_canon_sha256":"abfb85a78ed8dcffed88e09ed2b05ab13e798d77fba77565924cd420f1d98b4d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:48:51.157120Z","signature_b64":"goq4tUwWKzF46D5OllH5JapUlkJEj3jIrRKJ7d2p41Yix5Do1WvkJAXt2b2m5Vp5NspcvBczxB3ZdLVM+C8uAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"71b2f540d31488d4d92fc8b9baccf3b50b1e0c4e3b7fa55b410c8ec2c178c176","last_reissued_at":"2026-07-05T07:48:51.156747Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:48:51.156747Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Policy Optimization in RLHF: The Impact of Out-of-preference Data","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Tian Xu, Yang Yu, Ziniu Li","submitted_at":"2023-12-17T02:14:15Z","abstract_excerpt":"Aligning intelligent agents with human preferences and values is important. This paper examines two popular alignment methods: Direct Preference Optimization (DPO) and Reward-Model-Based Policy Optimization (RMB-PO). A variant of RMB-PO, referred to as RMB-PO+ is also considered. These methods, either explicitly or implicitly, learn a reward model from preference data and differ in the data used for policy optimization to unlock the generalization ability of the reward model. In particular, compared with DPO, RMB-PO additionally uses policy-generated data, and RMB-PO+ further leverages new, pr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.10584","kind":"arxiv","version":2},"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/2312.10584/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":"2312.10584","created_at":"2026-07-05T07:48:51.156803+00:00"},{"alias_kind":"arxiv_version","alias_value":"2312.10584v2","created_at":"2026-07-05T07:48:51.156803+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.10584","created_at":"2026-07-05T07:48:51.156803+00:00"},{"alias_kind":"pith_short_12","alias_value":"OGZPKQGTCSEN","created_at":"2026-07-05T07:48:51.156803+00:00"},{"alias_kind":"pith_short_16","alias_value":"OGZPKQGTCSENJWJP","created_at":"2026-07-05T07:48:51.156803+00:00"},{"alias_kind":"pith_short_8","alias_value":"OGZPKQGT","created_at":"2026-07-05T07:48:51.156803+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.20181","citing_title":"SGPO: Self-Generated Preference Optimization based on Self-Improver","ref_index":18,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OGZPKQGTCSENJWJPZC43VTHTWU","json":"https://pith.science/pith/OGZPKQGTCSENJWJPZC43VTHTWU.json","graph_json":"https://pith.science/api/pith-number/OGZPKQGTCSENJWJPZC43VTHTWU/graph.json","events_json":"https://pith.science/api/pith-number/OGZPKQGTCSENJWJPZC43VTHTWU/events.json","paper":"https://pith.science/paper/OGZPKQGT"},"agent_actions":{"view_html":"https://pith.science/pith/OGZPKQGTCSENJWJPZC43VTHTWU","download_json":"https://pith.science/pith/OGZPKQGTCSENJWJPZC43VTHTWU.json","view_paper":"https://pith.science/paper/OGZPKQGT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2312.10584&json=true","fetch_graph":"https://pith.science/api/pith-number/OGZPKQGTCSENJWJPZC43VTHTWU/graph.json","fetch_events":"https://pith.science/api/pith-number/OGZPKQGTCSENJWJPZC43VTHTWU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OGZPKQGTCSENJWJPZC43VTHTWU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OGZPKQGTCSENJWJPZC43VTHTWU/action/storage_attestation","attest_author":"https://pith.science/pith/OGZPKQGTCSENJWJPZC43VTHTWU/action/author_attestation","sign_citation":"https://pith.science/pith/OGZPKQGTCSENJWJPZC43VTHTWU/action/citation_signature","submit_replication":"https://pith.science/pith/OGZPKQGTCSENJWJPZC43VTHTWU/action/replication_record"}},"created_at":"2026-07-05T07:48:51.156803+00:00","updated_at":"2026-07-05T07:48:51.156803+00:00"}