{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:GLIPD37OQ23SYYLFHTLMTCM6LS","short_pith_number":"pith:GLIPD37O","schema_version":"1.0","canonical_sha256":"32d0f1efee86b72c61653cd6c9899e5c88c34bb50098d6f7a1f38aaf926c6ad4","source":{"kind":"arxiv","id":"2404.03715","version":1},"attestation_state":"computed","paper":{"title":"Direct Nash Optimization: Teaching Language Models to Self-Improve with General Preferences","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ahmed Awadallah, Arindam Mitra, Ching-An Cheng, Corby Rosset, Michael Santacroce, Tengyang Xie","submitted_at":"2024-04-04T17:56:41Z","abstract_excerpt":"This paper studies post-training large language models (LLMs) using preference feedback from a powerful oracle to help a model iteratively improve over itself. The typical approach for post-training LLMs involves Reinforcement Learning from Human Feedback (RLHF), which traditionally separates reward learning and subsequent policy optimization. However, such a reward maximization approach is limited by the nature of \"point-wise\" rewards (such as Bradley-Terry model), which fails to express complex intransitive or cyclic preference relations. While advances on RLHF show reward learning and polic"},"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":"2404.03715","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-04-04T17:56:41Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"447afa09a2785d3f245dab4ca48accd02be554af48c1dcc6a4630ffdd7e4f5f4","abstract_canon_sha256":"8b2a61351b37757790193007ca2c4d424a4718c88bef6a32ad4d55ef26930cb8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:04:40.584475Z","signature_b64":"DYBBa7WM4NtVa6DiBhDzrx0GRQQEKgbAxb4co0X58dxYETMDac1Yp5fU3q7Ro+WI1wvR/EdqFqHWV9Ohbm40Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"32d0f1efee86b72c61653cd6c9899e5c88c34bb50098d6f7a1f38aaf926c6ad4","last_reissued_at":"2026-07-05T08:04:40.583993Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:04:40.583993Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Direct Nash Optimization: Teaching Language Models to Self-Improve with General Preferences","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ahmed Awadallah, Arindam Mitra, Ching-An Cheng, Corby Rosset, Michael Santacroce, Tengyang Xie","submitted_at":"2024-04-04T17:56:41Z","abstract_excerpt":"This paper studies post-training large language models (LLMs) using preference feedback from a powerful oracle to help a model iteratively improve over itself. The typical approach for post-training LLMs involves Reinforcement Learning from Human Feedback (RLHF), which traditionally separates reward learning and subsequent policy optimization. However, such a reward maximization approach is limited by the nature of \"point-wise\" rewards (such as Bradley-Terry model), which fails to express complex intransitive or cyclic preference relations. While advances on RLHF show reward learning and polic"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.03715","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/2404.03715/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":"2404.03715","created_at":"2026-07-05T08:04:40.584045+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.03715v1","created_at":"2026-07-05T08:04:40.584045+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.03715","created_at":"2026-07-05T08:04:40.584045+00:00"},{"alias_kind":"pith_short_12","alias_value":"GLIPD37OQ23S","created_at":"2026-07-05T08:04:40.584045+00:00"},{"alias_kind":"pith_short_16","alias_value":"GLIPD37OQ23SYYLF","created_at":"2026-07-05T08:04:40.584045+00:00"},{"alias_kind":"pith_short_8","alias_value":"GLIPD37O","created_at":"2026-07-05T08:04:40.584045+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21550","citing_title":"AI Alignment From Social Choice Perspectives","ref_index":112,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03489","citing_title":"Learn from Your Mistakes: Tree-like Self-Play for Secure Code LLMs","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01561","citing_title":"S-SPPO: Semantic-Calibrated Self-Play Preference Optimization","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2504.12501","citing_title":"Reinforcement Learning from Human Feedback","ref_index":205,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15113","citing_title":"Learning from Language Feedback via Variational Policy Distillation","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17342","citing_title":"Transitivity Meets Cyclicity: Explicit Preference Decomposition for Dynamic Large Language Model Alignment","ref_index":91,"is_internal_anchor":false},{"citing_arxiv_id":"2509.23102","citing_title":"Multiplayer Nash Preference Optimization","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2409.12917","citing_title":"Training Language Models to Self-Correct via Reinforcement Learning","ref_index":90,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13875","citing_title":"Common-agency Games for Multi-Objective Test-Time Alignment","ref_index":63,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12288","citing_title":"TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching","ref_index":154,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12288","citing_title":"TokenRatio: Principled Token-Level Preference Optimization via Ratio Matching","ref_index":154,"is_internal_anchor":false},{"citing_arxiv_id":"2402.01306","citing_title":"KTO: Model Alignment as Prospect Theoretic Optimization","ref_index":15,"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":43,"is_internal_anchor":false},{"citing_arxiv_id":"2502.01456","citing_title":"Process Reinforcement through Implicit Rewards","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17501","citing_title":"CoAct: Co-Active LLM Preference Learning with Human-AI Synergy","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GLIPD37OQ23SYYLFHTLMTCM6LS","json":"https://pith.science/pith/GLIPD37OQ23SYYLFHTLMTCM6LS.json","graph_json":"https://pith.science/api/pith-number/GLIPD37OQ23SYYLFHTLMTCM6LS/graph.json","events_json":"https://pith.science/api/pith-number/GLIPD37OQ23SYYLFHTLMTCM6LS/events.json","paper":"https://pith.science/paper/GLIPD37O"},"agent_actions":{"view_html":"https://pith.science/pith/GLIPD37OQ23SYYLFHTLMTCM6LS","download_json":"https://pith.science/pith/GLIPD37OQ23SYYLFHTLMTCM6LS.json","view_paper":"https://pith.science/paper/GLIPD37O","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.03715&json=true","fetch_graph":"https://pith.science/api/pith-number/GLIPD37OQ23SYYLFHTLMTCM6LS/graph.json","fetch_events":"https://pith.science/api/pith-number/GLIPD37OQ23SYYLFHTLMTCM6LS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GLIPD37OQ23SYYLFHTLMTCM6LS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GLIPD37OQ23SYYLFHTLMTCM6LS/action/storage_attestation","attest_author":"https://pith.science/pith/GLIPD37OQ23SYYLFHTLMTCM6LS/action/author_attestation","sign_citation":"https://pith.science/pith/GLIPD37OQ23SYYLFHTLMTCM6LS/action/citation_signature","submit_replication":"https://pith.science/pith/GLIPD37OQ23SYYLFHTLMTCM6LS/action/replication_record"}},"created_at":"2026-07-05T08:04:40.584045+00:00","updated_at":"2026-07-05T08:04:40.584045+00:00"}