{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:6ZHGUD4IVKFOZ4URIG3WGW6CYB","short_pith_number":"pith:6ZHGUD4I","schema_version":"1.0","canonical_sha256":"f64e6a0f88aa8aecf29141b7635bc2c05112af5f4115bcb57caef0d99ce8aa16","source":{"kind":"arxiv","id":"2503.01864","version":1},"attestation_state":"computed","paper":{"title":"Larger or Smaller Reward Margins to Select Preferences for Alignment?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Bolin Ding, Jiancan Wu, Jinyang Gao, Junkang Wu, Kexin Huang, Xiangnan He, Xiang Wang, Xue Wang, Ziqian Chen","submitted_at":"2025-02-25T06:43:24Z","abstract_excerpt":"Preference learning is critical for aligning large language models (LLMs) with human values, with the quality of preference datasets playing a crucial role in this process. While existing metrics primarily assess data quality based on either explicit or implicit reward margins, they often provide contradictory evaluations for the same data. To address this issue, we introduce the alignment potential metric, which quantifies the gap from the model's current implicit reward margin to the target explicit reward margin, thereby estimating the model's potential to align with the preference data. Em"},"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":"2503.01864","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-02-25T06:43:24Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"b3fa748f3efb258d016871d5afe7877c00ea116778248f7b78a0fbd80b90d083","abstract_canon_sha256":"ad9551c81f2692311b7c2244e527f2959c0d428d18d29b21a5ee76b12ec7343a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:23:31.419150Z","signature_b64":"Iaa75D0sSRGW/kUjE2J595AG5bah8+rk93sdaaZmhP0IgqIyEQQ+6pYDlTE6hiFZDURjwSkNbqP7LjkN83v7BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f64e6a0f88aa8aecf29141b7635bc2c05112af5f4115bcb57caef0d99ce8aa16","last_reissued_at":"2026-07-05T10:23:31.418467Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:23:31.418467Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Larger or Smaller Reward Margins to Select Preferences for Alignment?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Bolin Ding, Jiancan Wu, Jinyang Gao, Junkang Wu, Kexin Huang, Xiangnan He, Xiang Wang, Xue Wang, Ziqian Chen","submitted_at":"2025-02-25T06:43:24Z","abstract_excerpt":"Preference learning is critical for aligning large language models (LLMs) with human values, with the quality of preference datasets playing a crucial role in this process. While existing metrics primarily assess data quality based on either explicit or implicit reward margins, they often provide contradictory evaluations for the same data. To address this issue, we introduce the alignment potential metric, which quantifies the gap from the model's current implicit reward margin to the target explicit reward margin, thereby estimating the model's potential to align with the preference data. Em"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.01864","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/2503.01864/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":"2503.01864","created_at":"2026-07-05T10:23:31.418547+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.01864v1","created_at":"2026-07-05T10:23:31.418547+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.01864","created_at":"2026-07-05T10:23:31.418547+00:00"},{"alias_kind":"pith_short_12","alias_value":"6ZHGUD4IVKFO","created_at":"2026-07-05T10:23:31.418547+00:00"},{"alias_kind":"pith_short_16","alias_value":"6ZHGUD4IVKFOZ4UR","created_at":"2026-07-05T10:23:31.418547+00:00"},{"alias_kind":"pith_short_8","alias_value":"6ZHGUD4I","created_at":"2026-07-05T10:23:31.418547+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2602.07340","citing_title":"Revisiting Robustness for LLM Safety Alignment via Selective Geometry Control","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/6ZHGUD4IVKFOZ4URIG3WGW6CYB","json":"https://pith.science/pith/6ZHGUD4IVKFOZ4URIG3WGW6CYB.json","graph_json":"https://pith.science/api/pith-number/6ZHGUD4IVKFOZ4URIG3WGW6CYB/graph.json","events_json":"https://pith.science/api/pith-number/6ZHGUD4IVKFOZ4URIG3WGW6CYB/events.json","paper":"https://pith.science/paper/6ZHGUD4I"},"agent_actions":{"view_html":"https://pith.science/pith/6ZHGUD4IVKFOZ4URIG3WGW6CYB","download_json":"https://pith.science/pith/6ZHGUD4IVKFOZ4URIG3WGW6CYB.json","view_paper":"https://pith.science/paper/6ZHGUD4I","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.01864&json=true","fetch_graph":"https://pith.science/api/pith-number/6ZHGUD4IVKFOZ4URIG3WGW6CYB/graph.json","fetch_events":"https://pith.science/api/pith-number/6ZHGUD4IVKFOZ4URIG3WGW6CYB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/6ZHGUD4IVKFOZ4URIG3WGW6CYB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/6ZHGUD4IVKFOZ4URIG3WGW6CYB/action/storage_attestation","attest_author":"https://pith.science/pith/6ZHGUD4IVKFOZ4URIG3WGW6CYB/action/author_attestation","sign_citation":"https://pith.science/pith/6ZHGUD4IVKFOZ4URIG3WGW6CYB/action/citation_signature","submit_replication":"https://pith.science/pith/6ZHGUD4IVKFOZ4URIG3WGW6CYB/action/replication_record"}},"created_at":"2026-07-05T10:23:31.418547+00:00","updated_at":"2026-07-05T10:23:31.418547+00:00"}