{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SPCWOQ7BUGJKTBI77APJ5Y5XUK","short_pith_number":"pith:SPCWOQ7B","schema_version":"1.0","canonical_sha256":"93c56743e1a192a9851ff81e9ee3b7a2b485db1d47a10060d0805b04a75ade0c","source":{"kind":"arxiv","id":"2502.00814","version":2},"attestation_state":"computed","paper":{"title":"Disentangling Length Bias In Preference Learning Via Response-Conditioned Modeling","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Houqiang Li, Jianfeng Cai, Jinhua Zhu, Li Li, Ruopei Sun, Wengang Zhou, Yue Wang","submitted_at":"2025-02-02T14:50:25Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has achieved considerable success in aligning large language models (LLMs) by modeling human preferences with a learnable reward model and employing a reinforcement learning algorithm to maximize the reward model's scores. However, these reward models are susceptible to exploitation through various superficial confounding factors, with length bias emerging as a particularly significant concern. Moreover, while the pronounced impact of length bias on preference modeling suggests that LLMs possess an inherent sensitivity to length perception, our"},"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":"2502.00814","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-02T14:50:25Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"f4eb5083672343c86a764accb285acaa4855c916de1d3734fa81b23e26d51a63","abstract_canon_sha256":"4288ca0e5f408eb0a4707a88f26873c458b51a4b6305a9f67d7b6e047ac30829"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:04:44.909817Z","signature_b64":"wBSOh2Hy2e+0xerIf515AqxPFf3uYmJuutmtNbk5JyAwKCaj56eL7xidzv6aB2YwUM0qTQiD67YnUoQMH8pWCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"93c56743e1a192a9851ff81e9ee3b7a2b485db1d47a10060d0805b04a75ade0c","last_reissued_at":"2026-07-05T11:04:44.909305Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:04:44.909305Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Disentangling Length Bias In Preference Learning Via Response-Conditioned Modeling","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Houqiang Li, Jianfeng Cai, Jinhua Zhu, Li Li, Ruopei Sun, Wengang Zhou, Yue Wang","submitted_at":"2025-02-02T14:50:25Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has achieved considerable success in aligning large language models (LLMs) by modeling human preferences with a learnable reward model and employing a reinforcement learning algorithm to maximize the reward model's scores. However, these reward models are susceptible to exploitation through various superficial confounding factors, with length bias emerging as a particularly significant concern. Moreover, while the pronounced impact of length bias on preference modeling suggests that LLMs possess an inherent sensitivity to length perception, our"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.00814","kind":"arxiv","version":2},"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/2502.00814/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":"2502.00814","created_at":"2026-07-05T11:04:44.909387+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.00814v2","created_at":"2026-07-05T11:04:44.909387+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.00814","created_at":"2026-07-05T11:04:44.909387+00:00"},{"alias_kind":"pith_short_12","alias_value":"SPCWOQ7BUGJK","created_at":"2026-07-05T11:04:44.909387+00:00"},{"alias_kind":"pith_short_16","alias_value":"SPCWOQ7BUGJKTBI7","created_at":"2026-07-05T11:04:44.909387+00:00"},{"alias_kind":"pith_short_8","alias_value":"SPCWOQ7B","created_at":"2026-07-05T11:04:44.909387+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2506.12382","citing_title":"Exploring the Secondary Risks of Large Language Models","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21911","citing_title":"When Prompts Override Vision: Prompt-Induced Hallucinations in LVLMs","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17328","citing_title":"Rethinking the Comparison Unit in Sequence-Level Reinforcement Learning: An Equal-Length Paired Training Framework from Loss Correction to Sample Construction","ref_index":17,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SPCWOQ7BUGJKTBI77APJ5Y5XUK","json":"https://pith.science/pith/SPCWOQ7BUGJKTBI77APJ5Y5XUK.json","graph_json":"https://pith.science/api/pith-number/SPCWOQ7BUGJKTBI77APJ5Y5XUK/graph.json","events_json":"https://pith.science/api/pith-number/SPCWOQ7BUGJKTBI77APJ5Y5XUK/events.json","paper":"https://pith.science/paper/SPCWOQ7B"},"agent_actions":{"view_html":"https://pith.science/pith/SPCWOQ7BUGJKTBI77APJ5Y5XUK","download_json":"https://pith.science/pith/SPCWOQ7BUGJKTBI77APJ5Y5XUK.json","view_paper":"https://pith.science/paper/SPCWOQ7B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.00814&json=true","fetch_graph":"https://pith.science/api/pith-number/SPCWOQ7BUGJKTBI77APJ5Y5XUK/graph.json","fetch_events":"https://pith.science/api/pith-number/SPCWOQ7BUGJKTBI77APJ5Y5XUK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SPCWOQ7BUGJKTBI77APJ5Y5XUK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SPCWOQ7BUGJKTBI77APJ5Y5XUK/action/storage_attestation","attest_author":"https://pith.science/pith/SPCWOQ7BUGJKTBI77APJ5Y5XUK/action/author_attestation","sign_citation":"https://pith.science/pith/SPCWOQ7BUGJKTBI77APJ5Y5XUK/action/citation_signature","submit_replication":"https://pith.science/pith/SPCWOQ7BUGJKTBI77APJ5Y5XUK/action/replication_record"}},"created_at":"2026-07-05T11:04:44.909387+00:00","updated_at":"2026-07-05T11:04:44.909387+00:00"}