{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:MDGCVOT5NN5CKOQ74P6EFKUACZ","short_pith_number":"pith:MDGCVOT5","schema_version":"1.0","canonical_sha256":"60cc2aba7d6b7a253a1fe3fc42aa801677fb05322b81788907502945d6751697","source":{"kind":"arxiv","id":"2507.01915","version":1},"attestation_state":"computed","paper":{"title":"Gradient-Adaptive Policy Optimization: Towards Multi-Objective Alignment of Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chengao Li, Hanyu Zhang, Hongyan Xue, Qing He, Xiang Ao, Yunkun Xu","submitted_at":"2025-07-02T17:25:26Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has emerged as a powerful technique for aligning large language models (LLMs) with human preferences. However, effectively aligning LLMs with diverse human preferences remains a significant challenge, particularly when they are conflict. To address this issue, we frame human value alignment as a multi-objective optimization problem, aiming to maximize a set of potentially conflicting objectives. We introduce Gradient-Adaptive Policy Optimization (GAPO), a novel fine-tuning paradigm that employs multiple-gradient descent to align LLMs with diver"},"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":"2507.01915","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-07-02T17:25:26Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"db3756dc78f3d608df2e487cdd6295ee2b56134e5b51f3c8d39c7bb68d64743e","abstract_canon_sha256":"40864f0267d246039297d5308208bed8b2ec161dc1e08d871a3755b3bc465b21"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:30:58.063311Z","signature_b64":"tWqpSSAMWYoduSDNBP+o3P+Ex5cS7CrVvNKxIJm8xKi3gj8jw8NzbSZlcE9dKPvwPVbYIzIH0j+eNwUhoYzNCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"60cc2aba7d6b7a253a1fe3fc42aa801677fb05322b81788907502945d6751697","last_reissued_at":"2026-07-05T11:30:58.062820Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:30:58.062820Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradient-Adaptive Policy Optimization: Towards Multi-Objective Alignment of Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chengao Li, Hanyu Zhang, Hongyan Xue, Qing He, Xiang Ao, Yunkun Xu","submitted_at":"2025-07-02T17:25:26Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has emerged as a powerful technique for aligning large language models (LLMs) with human preferences. However, effectively aligning LLMs with diverse human preferences remains a significant challenge, particularly when they are conflict. To address this issue, we frame human value alignment as a multi-objective optimization problem, aiming to maximize a set of potentially conflicting objectives. We introduce Gradient-Adaptive Policy Optimization (GAPO), a novel fine-tuning paradigm that employs multiple-gradient descent to align LLMs with diver"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.01915","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/2507.01915/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":"2507.01915","created_at":"2026-07-05T11:30:58.062882+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.01915v1","created_at":"2026-07-05T11:30:58.062882+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.01915","created_at":"2026-07-05T11:30:58.062882+00:00"},{"alias_kind":"pith_short_12","alias_value":"MDGCVOT5NN5C","created_at":"2026-07-05T11:30:58.062882+00:00"},{"alias_kind":"pith_short_16","alias_value":"MDGCVOT5NN5CKOQ7","created_at":"2026-07-05T11:30:58.062882+00:00"},{"alias_kind":"pith_short_8","alias_value":"MDGCVOT5","created_at":"2026-07-05T11:30:58.062882+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","ref_index":267,"is_internal_anchor":false},{"citing_arxiv_id":"2603.15646","citing_title":"Alternating Reinforcement Learning with Contextual Rubric Rewards: Beyond the Scalarization Strategy","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2603.18113","citing_title":"VC-Soup: Value-Consistency Guided Multi-Value Alignment for Large Language Models","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MDGCVOT5NN5CKOQ74P6EFKUACZ","json":"https://pith.science/pith/MDGCVOT5NN5CKOQ74P6EFKUACZ.json","graph_json":"https://pith.science/api/pith-number/MDGCVOT5NN5CKOQ74P6EFKUACZ/graph.json","events_json":"https://pith.science/api/pith-number/MDGCVOT5NN5CKOQ74P6EFKUACZ/events.json","paper":"https://pith.science/paper/MDGCVOT5"},"agent_actions":{"view_html":"https://pith.science/pith/MDGCVOT5NN5CKOQ74P6EFKUACZ","download_json":"https://pith.science/pith/MDGCVOT5NN5CKOQ74P6EFKUACZ.json","view_paper":"https://pith.science/paper/MDGCVOT5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.01915&json=true","fetch_graph":"https://pith.science/api/pith-number/MDGCVOT5NN5CKOQ74P6EFKUACZ/graph.json","fetch_events":"https://pith.science/api/pith-number/MDGCVOT5NN5CKOQ74P6EFKUACZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MDGCVOT5NN5CKOQ74P6EFKUACZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MDGCVOT5NN5CKOQ74P6EFKUACZ/action/storage_attestation","attest_author":"https://pith.science/pith/MDGCVOT5NN5CKOQ74P6EFKUACZ/action/author_attestation","sign_citation":"https://pith.science/pith/MDGCVOT5NN5CKOQ74P6EFKUACZ/action/citation_signature","submit_replication":"https://pith.science/pith/MDGCVOT5NN5CKOQ74P6EFKUACZ/action/replication_record"}},"created_at":"2026-07-05T11:30:58.062882+00:00","updated_at":"2026-07-05T11:30:58.062882+00:00"}