{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:O4UR3IOXQQFJLB6M2DYGAP6QHW","short_pith_number":"pith:O4UR3IOX","schema_version":"1.0","canonical_sha256":"77291da1d7840a9587ccd0f0603fd03d96df368267ddbd9352a7506d438adf2b","source":{"kind":"arxiv","id":"2405.00675","version":5},"attestation_state":"computed","paper":{"title":"Self-Play Preference Optimization for Language Model Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Huizhuo Yuan, Kaixuan Ji, Quanquan Gu, Yiming Yang, Yue Wu, Zhiqing Sun","submitted_at":"2024-05-01T17:59:20Z","abstract_excerpt":"Standard reinforcement learning from human feedback (RLHF) approaches relying on parametric models like the Bradley-Terry model fall short in capturing the intransitivity and irrationality in human preferences. Recent advancements suggest that directly working with preference probabilities can yield a more accurate reflection of human preferences, enabling more flexible and accurate language model alignment. In this paper, we propose a self-play-based method for language model alignment, which treats the problem as a constant-sum two-player game aimed at identifying the Nash equilibrium policy"},"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":"2405.00675","kind":"arxiv","version":5},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-05-01T17:59:20Z","cross_cats_sorted":["cs.AI","cs.CL","stat.ML"],"title_canon_sha256":"ae1180af663dafd07b80ea3c86f409cb2bfdd21191226a763ba7b51e9ed885c0","abstract_canon_sha256":"5effdfe3da4c7b9e483b953cb386385e7def5ad87ab67e3a84571e29ea7ace83"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:16:14.917567Z","signature_b64":"6C7RQrUx4DZZNNMUqdwhOex40LMk5GC/rtjM+iXRCaf5EDJzQdL15l61wuubtexvnmRIx6XgQPXqysSUZQR+Cw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"77291da1d7840a9587ccd0f0603fd03d96df368267ddbd9352a7506d438adf2b","last_reissued_at":"2026-07-05T09:16:14.917071Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:16:14.917071Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Self-Play Preference Optimization for Language Model Alignment","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Huizhuo Yuan, Kaixuan Ji, Quanquan Gu, Yiming Yang, Yue Wu, Zhiqing Sun","submitted_at":"2024-05-01T17:59:20Z","abstract_excerpt":"Standard reinforcement learning from human feedback (RLHF) approaches relying on parametric models like the Bradley-Terry model fall short in capturing the intransitivity and irrationality in human preferences. Recent advancements suggest that directly working with preference probabilities can yield a more accurate reflection of human preferences, enabling more flexible and accurate language model alignment. In this paper, we propose a self-play-based method for language model alignment, which treats the problem as a constant-sum two-player game aimed at identifying the Nash equilibrium policy"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.00675","kind":"arxiv","version":5},"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/2405.00675/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":"2405.00675","created_at":"2026-07-05T09:16:14.917136+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.00675v5","created_at":"2026-07-05T09:16:14.917136+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.00675","created_at":"2026-07-05T09:16:14.917136+00:00"},{"alias_kind":"pith_short_12","alias_value":"O4UR3IOXQQFJ","created_at":"2026-07-05T09:16:14.917136+00:00"},{"alias_kind":"pith_short_16","alias_value":"O4UR3IOXQQFJLB6M","created_at":"2026-07-05T09:16:14.917136+00:00"},{"alias_kind":"pith_short_8","alias_value":"O4UR3IOX","created_at":"2026-07-05T09:16:14.917136+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":18,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.28401","citing_title":"Vision-driven Preference Synthesis for Mitigating Hallucinations in VLMs","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28440","citing_title":"AdaDPO: Self-Adaptive Direct Preference Optimization with Balanced Gradient Updates","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29398","citing_title":"GDSD: Reinforcement Learning as Guided Denoiser Self-Distillation for Diffusion Language Models","ref_index":57,"is_internal_anchor":false},{"citing_arxiv_id":"2408.15339","citing_title":"UNA: A Unified Supervised Framework for Efficient LLM Alignment Across Feedback Types","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18721","citing_title":"General Preference Reinforcement Learning","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18721","citing_title":"General Preference Reinforcement Learning","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15113","citing_title":"Learning from Language Feedback via Variational Policy Distillation","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18721","citing_title":"General Preference Reinforcement Learning","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2509.23102","citing_title":"Multiplayer Nash Preference Optimization","ref_index":31,"is_internal_anchor":false},{"citing_arxiv_id":"2411.10442","citing_title":"Enhancing the Reasoning Ability of Multimodal Large Language Models via Mixed Preference Optimization","ref_index":105,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09363","citing_title":"Near-Optimal Last-Iterate Convergence for Zero-Sum Games with Bandit Feedback and Opponent Actions","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09214","citing_title":"Fast Rates for Offline Contextual Bandits with Forward-KL Regularization under Single-Policy Concentrability","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09922","citing_title":"Team-Based Self-Play With Dual Adaptive Weighting for Fine-Tuning LLMs","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04180","citing_title":"MedFabric and EtHER: A Data-Centric Framework for Word-Level Fabrication Generation and Detection in Medical LLMs","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2604.20933","citing_title":"IRIS: Interpolative R\\'enyi Iterative Self-play for Large Language Model Fine-Tuning","ref_index":81,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07331","citing_title":"Rethinking Importance Sampling in LLM Policy Optimization: A Cumulative Token Perspective","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07244","citing_title":"Experience Sharing in Mutual Reinforcement Learning for Heterogeneous Language Models","ref_index":97,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02626","citing_title":"Gradient-Gated DPO: Stabilizing Preference Optimization in Language Models","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/O4UR3IOXQQFJLB6M2DYGAP6QHW","json":"https://pith.science/pith/O4UR3IOXQQFJLB6M2DYGAP6QHW.json","graph_json":"https://pith.science/api/pith-number/O4UR3IOXQQFJLB6M2DYGAP6QHW/graph.json","events_json":"https://pith.science/api/pith-number/O4UR3IOXQQFJLB6M2DYGAP6QHW/events.json","paper":"https://pith.science/paper/O4UR3IOX"},"agent_actions":{"view_html":"https://pith.science/pith/O4UR3IOXQQFJLB6M2DYGAP6QHW","download_json":"https://pith.science/pith/O4UR3IOXQQFJLB6M2DYGAP6QHW.json","view_paper":"https://pith.science/paper/O4UR3IOX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.00675&json=true","fetch_graph":"https://pith.science/api/pith-number/O4UR3IOXQQFJLB6M2DYGAP6QHW/graph.json","fetch_events":"https://pith.science/api/pith-number/O4UR3IOXQQFJLB6M2DYGAP6QHW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/O4UR3IOXQQFJLB6M2DYGAP6QHW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/O4UR3IOXQQFJLB6M2DYGAP6QHW/action/storage_attestation","attest_author":"https://pith.science/pith/O4UR3IOXQQFJLB6M2DYGAP6QHW/action/author_attestation","sign_citation":"https://pith.science/pith/O4UR3IOXQQFJLB6M2DYGAP6QHW/action/citation_signature","submit_replication":"https://pith.science/pith/O4UR3IOXQQFJLB6M2DYGAP6QHW/action/replication_record"}},"created_at":"2026-07-05T09:16:14.917136+00:00","updated_at":"2026-07-05T09:16:14.917136+00:00"}