{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SGNGAQJ62EBBZ5GOXAJ4Z7K33J","short_pith_number":"pith:SGNGAQJ6","schema_version":"1.0","canonical_sha256":"919a60413ed1021cf4ceb813ccfd5bda4f4fde2eada897c6b612f4d0b8cb9cef","source":{"kind":"arxiv","id":"2501.02790","version":1},"attestation_state":"computed","paper":{"title":"Segmenting Text and Learning Their Rewards for Improved RLHF in Language Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Hany Awadalla, Mingyuan Zhou, Shentao Yang, Weizhu Chen, Yueqin Yin, Yujia Xie, Yuting Sun, Ziyi Yang","submitted_at":"2025-01-06T06:17:56Z","abstract_excerpt":"Reinforcement learning from human feedback (RLHF) has been widely adopted to align language models (LMs) with human preference. Prior RLHF works typically take a bandit formulation, which, though intuitive, ignores the sequential nature of LM generation and can suffer from the sparse reward issue. While recent works propose dense token-level RLHF, treating each token as an action may be oversubtle to proper reward assignment. In this paper, we seek to get the best of both by training and utilizing a segment-level reward model, which assigns a reward to each semantically complete text segment t"},"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":"2501.02790","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-01-06T06:17:56Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"87e1bd669244c78947c89f0a9e1bcda26b6c7cf3e909b5030129a4a3aec8fa4b","abstract_canon_sha256":"52f22efbb3828268e197985c4080609384c3416143399d03b95b1f09049bf555"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:57:12.270959Z","signature_b64":"0QSuQ8OTAAoIdoqw0n++276KYKGkn5jOw68mORkmInp+9rgn1X04HLUjyWunaG2Gb8Y7cESL+0ql4F3DjOdjDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"919a60413ed1021cf4ceb813ccfd5bda4f4fde2eada897c6b612f4d0b8cb9cef","last_reissued_at":"2026-07-05T09:57:12.270516Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:57:12.270516Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Segmenting Text and Learning Their Rewards for Improved RLHF in Language Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Hany Awadalla, Mingyuan Zhou, Shentao Yang, Weizhu Chen, Yueqin Yin, Yujia Xie, Yuting Sun, Ziyi Yang","submitted_at":"2025-01-06T06:17:56Z","abstract_excerpt":"Reinforcement learning from human feedback (RLHF) has been widely adopted to align language models (LMs) with human preference. Prior RLHF works typically take a bandit formulation, which, though intuitive, ignores the sequential nature of LM generation and can suffer from the sparse reward issue. While recent works propose dense token-level RLHF, treating each token as an action may be oversubtle to proper reward assignment. In this paper, we seek to get the best of both by training and utilizing a segment-level reward model, which assigns a reward to each semantically complete text segment t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.02790","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/2501.02790/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":"2501.02790","created_at":"2026-07-05T09:57:12.270575+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.02790v1","created_at":"2026-07-05T09:57:12.270575+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.02790","created_at":"2026-07-05T09:57:12.270575+00:00"},{"alias_kind":"pith_short_12","alias_value":"SGNGAQJ62EBB","created_at":"2026-07-05T09:57:12.270575+00:00"},{"alias_kind":"pith_short_16","alias_value":"SGNGAQJ62EBBZ5GO","created_at":"2026-07-05T09:57:12.270575+00:00"},{"alias_kind":"pith_short_8","alias_value":"SGNGAQJ6","created_at":"2026-07-05T09:57:12.270575+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09711","citing_title":"Proxy Reward Internalization and Mechanistic Exploitation: A Learned Precursor to Reward Hacking and Its Generalization","ref_index":151,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13602","citing_title":"Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges","ref_index":117,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21160","citing_title":"Reinforcing 3D Understanding in Point-VLMs via Geometric Reward Credit Assignment","ref_index":41,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SGNGAQJ62EBBZ5GOXAJ4Z7K33J","json":"https://pith.science/pith/SGNGAQJ62EBBZ5GOXAJ4Z7K33J.json","graph_json":"https://pith.science/api/pith-number/SGNGAQJ62EBBZ5GOXAJ4Z7K33J/graph.json","events_json":"https://pith.science/api/pith-number/SGNGAQJ62EBBZ5GOXAJ4Z7K33J/events.json","paper":"https://pith.science/paper/SGNGAQJ6"},"agent_actions":{"view_html":"https://pith.science/pith/SGNGAQJ62EBBZ5GOXAJ4Z7K33J","download_json":"https://pith.science/pith/SGNGAQJ62EBBZ5GOXAJ4Z7K33J.json","view_paper":"https://pith.science/paper/SGNGAQJ6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.02790&json=true","fetch_graph":"https://pith.science/api/pith-number/SGNGAQJ62EBBZ5GOXAJ4Z7K33J/graph.json","fetch_events":"https://pith.science/api/pith-number/SGNGAQJ62EBBZ5GOXAJ4Z7K33J/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SGNGAQJ62EBBZ5GOXAJ4Z7K33J/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SGNGAQJ62EBBZ5GOXAJ4Z7K33J/action/storage_attestation","attest_author":"https://pith.science/pith/SGNGAQJ62EBBZ5GOXAJ4Z7K33J/action/author_attestation","sign_citation":"https://pith.science/pith/SGNGAQJ62EBBZ5GOXAJ4Z7K33J/action/citation_signature","submit_replication":"https://pith.science/pith/SGNGAQJ62EBBZ5GOXAJ4Z7K33J/action/replication_record"}},"created_at":"2026-07-05T09:57:12.270575+00:00","updated_at":"2026-07-05T09:57:12.270575+00:00"}