{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:4UOHAOSX5FZ6C7HQGEFABGPXYE","short_pith_number":"pith:4UOHAOSX","schema_version":"1.0","canonical_sha256":"e51c703a57e973e17cf0310a0099f7c12d7144532527c649040005ba927cdca4","source":{"kind":"arxiv","id":"2410.05584","version":5},"attestation_state":"computed","paper":{"title":"Rethinking Reward Model Evaluation: Are We Barking up the Wrong Tree?","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ben He, Debing Zhang, Hongyu Lin, Jie Lou, Le Sun, Xianpei Han, Xing Yu, Xinyu Lu, Xueru Wen, Yaojie Lu","submitted_at":"2024-10-08T00:52:03Z","abstract_excerpt":"Reward Models (RMs) are crucial for aligning language models with human preferences. Currently, the evaluation of RMs depends on measuring accuracy against a validation set of manually annotated preference data. Although this method is straightforward and widely adopted, the relationship between RM accuracy and downstream policy performance remains under-explored. In this work, we conduct experiments in a synthetic setting to investigate how differences in RM measured by accuracy translate into gaps in optimized policy performance. Our findings reveal that while there is a weak positive correl"},"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":"2410.05584","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-08T00:52:03Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"4eba20eca6d44e2a2df8be09acf4632b68b2e6c3ca89693603506f4f1d7a1be5","abstract_canon_sha256":"dc140aab1452e1f759861c55dcb53db270263e6fd1436c04097620cf1627ec39"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:14:06.121645Z","signature_b64":"H7Lfsw6RIDNuVOJ1fMDYFpNMVlmc40FxYwtrjMOThi0VUL3jlAvurC4hmJC9pxgkQTHAAHvCbDJnu9XUMpKyBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e51c703a57e973e17cf0310a0099f7c12d7144532527c649040005ba927cdca4","last_reissued_at":"2026-07-05T10:14:06.121143Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:14:06.121143Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rethinking Reward Model Evaluation: Are We Barking up the Wrong Tree?","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Ben He, Debing Zhang, Hongyu Lin, Jie Lou, Le Sun, Xianpei Han, Xing Yu, Xinyu Lu, Xueru Wen, Yaojie Lu","submitted_at":"2024-10-08T00:52:03Z","abstract_excerpt":"Reward Models (RMs) are crucial for aligning language models with human preferences. Currently, the evaluation of RMs depends on measuring accuracy against a validation set of manually annotated preference data. Although this method is straightforward and widely adopted, the relationship between RM accuracy and downstream policy performance remains under-explored. In this work, we conduct experiments in a synthetic setting to investigate how differences in RM measured by accuracy translate into gaps in optimized policy performance. Our findings reveal that while there is a weak positive correl"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.05584","kind":"arxiv","version":5},"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/2410.05584/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":"2410.05584","created_at":"2026-07-05T10:14:06.121202+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.05584v5","created_at":"2026-07-05T10:14:06.121202+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.05584","created_at":"2026-07-05T10:14:06.121202+00:00"},{"alias_kind":"pith_short_12","alias_value":"4UOHAOSX5FZ6","created_at":"2026-07-05T10:14:06.121202+00:00"},{"alias_kind":"pith_short_16","alias_value":"4UOHAOSX5FZ6C7HQ","created_at":"2026-07-05T10:14:06.121202+00:00"},{"alias_kind":"pith_short_8","alias_value":"4UOHAOSX","created_at":"2026-07-05T10:14:06.121202+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2506.01937","citing_title":"RewardBench 2: Advancing Reward Model Evaluation","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14539","citing_title":"Learning from Failures: Correction-Oriented Policy Optimization with Verifiable Rewards","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07343","citing_title":"Personalized RewardBench: Evaluating Reward Models with Human Aligned Personalization","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18327","citing_title":"PARM: Pipeline-Adapted Reward Model","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4UOHAOSX5FZ6C7HQGEFABGPXYE","json":"https://pith.science/pith/4UOHAOSX5FZ6C7HQGEFABGPXYE.json","graph_json":"https://pith.science/api/pith-number/4UOHAOSX5FZ6C7HQGEFABGPXYE/graph.json","events_json":"https://pith.science/api/pith-number/4UOHAOSX5FZ6C7HQGEFABGPXYE/events.json","paper":"https://pith.science/paper/4UOHAOSX"},"agent_actions":{"view_html":"https://pith.science/pith/4UOHAOSX5FZ6C7HQGEFABGPXYE","download_json":"https://pith.science/pith/4UOHAOSX5FZ6C7HQGEFABGPXYE.json","view_paper":"https://pith.science/paper/4UOHAOSX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.05584&json=true","fetch_graph":"https://pith.science/api/pith-number/4UOHAOSX5FZ6C7HQGEFABGPXYE/graph.json","fetch_events":"https://pith.science/api/pith-number/4UOHAOSX5FZ6C7HQGEFABGPXYE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4UOHAOSX5FZ6C7HQGEFABGPXYE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4UOHAOSX5FZ6C7HQGEFABGPXYE/action/storage_attestation","attest_author":"https://pith.science/pith/4UOHAOSX5FZ6C7HQGEFABGPXYE/action/author_attestation","sign_citation":"https://pith.science/pith/4UOHAOSX5FZ6C7HQGEFABGPXYE/action/citation_signature","submit_replication":"https://pith.science/pith/4UOHAOSX5FZ6C7HQGEFABGPXYE/action/replication_record"}},"created_at":"2026-07-05T10:14:06.121202+00:00","updated_at":"2026-07-05T10:14:06.121202+00:00"}