{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:LLEWARJRKY2K7UVNKZDU556HEW","short_pith_number":"pith:LLEWARJR","schema_version":"1.0","canonical_sha256":"5ac96045315634afd2ad56474ef7c725b451da32112f12cc9dea1344d2b2b884","source":{"kind":"arxiv","id":"2505.16265","version":1},"attestation_state":"computed","paper":{"title":"Think-RM: Enabling Long-Horizon Reasoning in Generative Reward Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Changlong Yu, Chao Zhang, Haoming Jiang, Ilgee Hong, Liang Qiu, Qingru Zhang, Qin Lu, Tuo Zhao, Weixiang Yan, Xin Liu, Zhenghao Xu","submitted_at":"2025-05-22T05:56:11Z","abstract_excerpt":"Reinforcement learning from human feedback (RLHF) has become a powerful post-training paradigm for aligning large language models with human preferences. A core challenge in RLHF is constructing accurate reward signals, where the conventional Bradley-Terry reward models (BT RMs) often suffer from sensitivity to data size and coverage, as well as vulnerability to reward hacking. Generative reward models (GenRMs) offer a more robust alternative by generating chain-of-thought (CoT) rationales followed by a final reward. However, existing GenRMs rely on shallow, vertically scaled reasoning, limiti"},"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":"2505.16265","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-22T05:56:11Z","cross_cats_sorted":[],"title_canon_sha256":"a0d5368683101b468cb2ce4f94c2cc0a5b691a46e4c141bae1f3d0d70d9c1034","abstract_canon_sha256":"a40ad3433cc6a2d6032e7eda88fff3aa74e409dc702b056230cf62e27b8d42fe"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:07:32.431929Z","signature_b64":"E6G/H3XAkalB9g78WG61eWFW6rpBl1YUBz824hNG9ANHBGGhOtRPy7B5XspivbPpzqcELjZ1bOmzFCDB4PJOBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5ac96045315634afd2ad56474ef7c725b451da32112f12cc9dea1344d2b2b884","last_reissued_at":"2026-07-05T11:07:32.431513Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:07:32.431513Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Think-RM: Enabling Long-Horizon Reasoning in Generative Reward Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Changlong Yu, Chao Zhang, Haoming Jiang, Ilgee Hong, Liang Qiu, Qingru Zhang, Qin Lu, Tuo Zhao, Weixiang Yan, Xin Liu, Zhenghao Xu","submitted_at":"2025-05-22T05:56:11Z","abstract_excerpt":"Reinforcement learning from human feedback (RLHF) has become a powerful post-training paradigm for aligning large language models with human preferences. A core challenge in RLHF is constructing accurate reward signals, where the conventional Bradley-Terry reward models (BT RMs) often suffer from sensitivity to data size and coverage, as well as vulnerability to reward hacking. Generative reward models (GenRMs) offer a more robust alternative by generating chain-of-thought (CoT) rationales followed by a final reward. However, existing GenRMs rely on shallow, vertically scaled reasoning, limiti"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.16265","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/2505.16265/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":"2505.16265","created_at":"2026-07-05T11:07:32.431571+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.16265v1","created_at":"2026-07-05T11:07:32.431571+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.16265","created_at":"2026-07-05T11:07:32.431571+00:00"},{"alias_kind":"pith_short_12","alias_value":"LLEWARJRKY2K","created_at":"2026-07-05T11:07:32.431571+00:00"},{"alias_kind":"pith_short_16","alias_value":"LLEWARJRKY2K7UVN","created_at":"2026-07-05T11:07:32.431571+00:00"},{"alias_kind":"pith_short_8","alias_value":"LLEWARJR","created_at":"2026-07-05T11:07:32.431571+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.01249","citing_title":"Trust Region On-Policy Distillation","ref_index":153,"is_internal_anchor":false},{"citing_arxiv_id":"2508.03556","citing_title":"VRPRM: Process Reward Modeling via Visual Reasoning","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27505","citing_title":"Leveraging Verifier-Based Reinforcement Learning in Image Editing","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2509.02547","citing_title":"The Landscape of Agentic Reinforcement Learning for LLMs: A Survey","ref_index":266,"is_internal_anchor":false},{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","ref_index":196,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27505","citing_title":"Leveraging Verifier-Based Reinforcement Learning in Image Editing","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LLEWARJRKY2K7UVNKZDU556HEW","json":"https://pith.science/pith/LLEWARJRKY2K7UVNKZDU556HEW.json","graph_json":"https://pith.science/api/pith-number/LLEWARJRKY2K7UVNKZDU556HEW/graph.json","events_json":"https://pith.science/api/pith-number/LLEWARJRKY2K7UVNKZDU556HEW/events.json","paper":"https://pith.science/paper/LLEWARJR"},"agent_actions":{"view_html":"https://pith.science/pith/LLEWARJRKY2K7UVNKZDU556HEW","download_json":"https://pith.science/pith/LLEWARJRKY2K7UVNKZDU556HEW.json","view_paper":"https://pith.science/paper/LLEWARJR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.16265&json=true","fetch_graph":"https://pith.science/api/pith-number/LLEWARJRKY2K7UVNKZDU556HEW/graph.json","fetch_events":"https://pith.science/api/pith-number/LLEWARJRKY2K7UVNKZDU556HEW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LLEWARJRKY2K7UVNKZDU556HEW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LLEWARJRKY2K7UVNKZDU556HEW/action/storage_attestation","attest_author":"https://pith.science/pith/LLEWARJRKY2K7UVNKZDU556HEW/action/author_attestation","sign_citation":"https://pith.science/pith/LLEWARJRKY2K7UVNKZDU556HEW/action/citation_signature","submit_replication":"https://pith.science/pith/LLEWARJRKY2K7UVNKZDU556HEW/action/replication_record"}},"created_at":"2026-07-05T11:07:32.431571+00:00","updated_at":"2026-07-05T11:07:32.431571+00:00"}