{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:EMX4ERL7KEIMZVWXHSMI2MQL3X","short_pith_number":"pith:EMX4ERL7","schema_version":"1.0","canonical_sha256":"232fc2457f5110ccd6d73c988d320bddd66799dbbfd5d648880c39758130bcdb","source":{"kind":"arxiv","id":"2403.19159","version":2},"attestation_state":"computed","paper":{"title":"Disentangling Length from Quality in Direct Preference Optimization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Chelsea Finn, Rafael Rafailov, Ryan Park, Stefano Ermon","submitted_at":"2024-03-28T06:03:47Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has been a crucial component in the recent success of Large Language Models. However, RLHF is know to exploit biases in human preferences, such as verbosity. A well-formatted and eloquent answer is often more highly rated by users, even when it is less helpful and objective. A number of approaches have been developed to control those biases in the classical RLHF literature, but the problem remains relatively under-explored for Direct Alignment Algorithms such as Direct Preference Optimization (DPO). Unlike classical RLHF, DPO does not train a s"},"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":"2403.19159","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-03-28T06:03:47Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"ebc0df157ee15775030357d9ced5924cfc14d10983aa0c0d39eff49d4f590850","abstract_canon_sha256":"1b0e6fb56f2466560089af4054cf9eb1d31d6ec4d03ee7fb5a5d7177817b128b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:04:34.094672Z","signature_b64":"XkJ0BcxZTrXGJJWMWnld+LxPjRKpD6JeDPFXyKu+7PDnPxLAe2W8vs0GFsl2lhBtbPgEEQ4YNoS8aPJhc3PbDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"232fc2457f5110ccd6d73c988d320bddd66799dbbfd5d648880c39758130bcdb","last_reissued_at":"2026-07-05T09:04:34.094134Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:04:34.094134Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Disentangling Length from Quality in Direct Preference Optimization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Chelsea Finn, Rafael Rafailov, Ryan Park, Stefano Ermon","submitted_at":"2024-03-28T06:03:47Z","abstract_excerpt":"Reinforcement Learning from Human Feedback (RLHF) has been a crucial component in the recent success of Large Language Models. However, RLHF is know to exploit biases in human preferences, such as verbosity. A well-formatted and eloquent answer is often more highly rated by users, even when it is less helpful and objective. A number of approaches have been developed to control those biases in the classical RLHF literature, but the problem remains relatively under-explored for Direct Alignment Algorithms such as Direct Preference Optimization (DPO). Unlike classical RLHF, DPO does not train a s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.19159","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/2403.19159/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":"2403.19159","created_at":"2026-07-05T09:04:34.094189+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.19159v2","created_at":"2026-07-05T09:04:34.094189+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.19159","created_at":"2026-07-05T09:04:34.094189+00:00"},{"alias_kind":"pith_short_12","alias_value":"EMX4ERL7KEIM","created_at":"2026-07-05T09:04:34.094189+00:00"},{"alias_kind":"pith_short_16","alias_value":"EMX4ERL7KEIMZVWX","created_at":"2026-07-05T09:04:34.094189+00:00"},{"alias_kind":"pith_short_8","alias_value":"EMX4ERL7","created_at":"2026-07-05T09:04:34.094189+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25432","citing_title":"Brevity is the Soul of Inference Efficiency: Inducing Concision in VLMs via Data Curation","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2606.25432","citing_title":"Brevity is the Soul of Inference Efficiency: Inducing Concision in VLMs via Data Curation","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28440","citing_title":"AdaDPO: Self-Adaptive Direct Preference Optimization with Balanced Gradient Updates","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2410.19471","citing_title":"Improving Inverse Folding for Peptide Design with Diversity-regularized Direct Preference Optimization","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2508.04149","citing_title":"Difficulty-Based Preference Data Selection by DPO Implicit Reward Gap","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2601.21350","citing_title":"Factored Causal Representation Learning for Robust Reward Modeling in RLHF","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2509.23102","citing_title":"Multiplayer Nash Preference Optimization","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04539","citing_title":"RLearner-LLM: Balancing Logical Grounding and Fluency in Large Language Models via Hybrid Direct Preference Optimization","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11134","citing_title":"Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08378","citing_title":"Reinforcement Learning for Scalable and Trustworthy Intelligent Systems","ref_index":101,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04539","citing_title":"RLearner-LLM: Balancing Logical Grounding and Fluency in Large Language Models via Hybrid Direct Preference Optimization","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04539","citing_title":"RLearner-LLM: Balancing Logical Grounding and Fluency in Large Language Models via Hybrid Direct Preference Optimization","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19016","citing_title":"AlignCultura: Towards Culturally Aligned Large Language Models?","ref_index":74,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06987","citing_title":"Response Time Enhances Alignment with Heterogeneous Preferences","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02626","citing_title":"Gradient-Gated DPO: Stabilizing Preference Optimization in Language Models","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EMX4ERL7KEIMZVWXHSMI2MQL3X","json":"https://pith.science/pith/EMX4ERL7KEIMZVWXHSMI2MQL3X.json","graph_json":"https://pith.science/api/pith-number/EMX4ERL7KEIMZVWXHSMI2MQL3X/graph.json","events_json":"https://pith.science/api/pith-number/EMX4ERL7KEIMZVWXHSMI2MQL3X/events.json","paper":"https://pith.science/paper/EMX4ERL7"},"agent_actions":{"view_html":"https://pith.science/pith/EMX4ERL7KEIMZVWXHSMI2MQL3X","download_json":"https://pith.science/pith/EMX4ERL7KEIMZVWXHSMI2MQL3X.json","view_paper":"https://pith.science/paper/EMX4ERL7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.19159&json=true","fetch_graph":"https://pith.science/api/pith-number/EMX4ERL7KEIMZVWXHSMI2MQL3X/graph.json","fetch_events":"https://pith.science/api/pith-number/EMX4ERL7KEIMZVWXHSMI2MQL3X/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EMX4ERL7KEIMZVWXHSMI2MQL3X/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EMX4ERL7KEIMZVWXHSMI2MQL3X/action/storage_attestation","attest_author":"https://pith.science/pith/EMX4ERL7KEIMZVWXHSMI2MQL3X/action/author_attestation","sign_citation":"https://pith.science/pith/EMX4ERL7KEIMZVWXHSMI2MQL3X/action/citation_signature","submit_replication":"https://pith.science/pith/EMX4ERL7KEIMZVWXHSMI2MQL3X/action/replication_record"}},"created_at":"2026-07-05T09:04:34.094189+00:00","updated_at":"2026-07-05T09:04:34.094189+00:00"}