{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:DEEMQ3QRH6D7KQSWY46TRF74AB","short_pith_number":"pith:DEEMQ3QR","schema_version":"1.0","canonical_sha256":"1908c86e113f87f54256c73d3897fc006d31b4eb43b9346c104577fbcb773857","source":{"kind":"arxiv","id":"2410.16184","version":1},"attestation_state":"computed","paper":{"title":"RM-Bench: Benchmarking Reward Models of Language Models with Subtlety and Style","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Juanzi Li, Lei Hou, Rui Min, Yantao Liu, Yixin Cao, Zijun Yao","submitted_at":"2024-10-21T16:48:26Z","abstract_excerpt":"Reward models are critical in techniques like Reinforcement Learning from Human Feedback (RLHF) and Inference Scaling Laws, where they guide language model alignment and select optimal responses. Despite their importance, existing reward model benchmarks often evaluate models by asking them to distinguish between responses generated by models of varying power. However, this approach fails to assess reward models on subtle but critical content changes and variations in style, resulting in a low correlation with policy model performance. To this end, we introduce RM-Bench, a novel benchmark desi"},"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":"2410.16184","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-21T16:48:26Z","cross_cats_sorted":[],"title_canon_sha256":"d3749f6d8253bde1c093357c039be96f82bc423a8a0cdc8ef149b94d82242c86","abstract_canon_sha256":"2aa97bd82670895d8bef30c27243115d74eac6e2cfa09638b03b7f9ce67bc99e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:23:31.001597Z","signature_b64":"s06HxrsC7KHYZVPwJYT8UvuJd/pWa3qMy+lXP/FR8sK+kDt4e4fdc+vDonCVsv0BYwnMgPFc4owGJ166J++fAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1908c86e113f87f54256c73d3897fc006d31b4eb43b9346c104577fbcb773857","last_reissued_at":"2026-07-05T09:23:31.001056Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:23:31.001056Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RM-Bench: Benchmarking Reward Models of Language Models with Subtlety and Style","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Juanzi Li, Lei Hou, Rui Min, Yantao Liu, Yixin Cao, Zijun Yao","submitted_at":"2024-10-21T16:48:26Z","abstract_excerpt":"Reward models are critical in techniques like Reinforcement Learning from Human Feedback (RLHF) and Inference Scaling Laws, where they guide language model alignment and select optimal responses. Despite their importance, existing reward model benchmarks often evaluate models by asking them to distinguish between responses generated by models of varying power. However, this approach fails to assess reward models on subtle but critical content changes and variations in style, resulting in a low correlation with policy model performance. To this end, we introduce RM-Bench, a novel benchmark desi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.16184","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/2410.16184/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":"2410.16184","created_at":"2026-07-05T09:23:31.001120+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.16184v1","created_at":"2026-07-05T09:23:31.001120+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.16184","created_at":"2026-07-05T09:23:31.001120+00:00"},{"alias_kind":"pith_short_12","alias_value":"DEEMQ3QRH6D7","created_at":"2026-07-05T09:23:31.001120+00:00"},{"alias_kind":"pith_short_16","alias_value":"DEEMQ3QRH6D7KQSW","created_at":"2026-07-05T09:23:31.001120+00:00"},{"alias_kind":"pith_short_8","alias_value":"DEEMQ3QR","created_at":"2026-07-05T09:23:31.001120+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25996","citing_title":"Autodata: An agentic data scientist to create high quality synthetic data","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2606.25996","citing_title":"Autodata: An agentic data scientist to create high quality synthetic data","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2606.32034","citing_title":"QVal: Cheaply Evaluating Dense Supervision Signals for Long-Horizon LLM Agents","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00609","citing_title":"CARE-RL: Capability-Aware Reinforcement Learning for Mitigating Cross-Domain Conflicts","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2509.21319","citing_title":"RLBFF: Binary Flexible Feedback to bridge between Human Feedback & Verifiable Rewards","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2506.01937","citing_title":"RewardBench 2: Advancing Reward Model Evaluation","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2512.20856","citing_title":"NVIDIA Nemotron 3: Efficient and Open Intelligence","ref_index":183,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12384","citing_title":"Scalable Token-Level Hallucination Detection in Large Language Models","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10805","citing_title":"Reasoning Is Not Free: Robust Adaptive Cost-Efficient Routing for LLM-as-a-Judge","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2412.05579","citing_title":"LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methods","ref_index":153,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07872","citing_title":"Video Understanding Reward Modeling: A Robust Benchmark and Performant Reward Models","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17282","citing_title":"MedPRMBench: A Fine-grained Benchmark for Process Reward Models in Medical Reasoning","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DEEMQ3QRH6D7KQSWY46TRF74AB","json":"https://pith.science/pith/DEEMQ3QRH6D7KQSWY46TRF74AB.json","graph_json":"https://pith.science/api/pith-number/DEEMQ3QRH6D7KQSWY46TRF74AB/graph.json","events_json":"https://pith.science/api/pith-number/DEEMQ3QRH6D7KQSWY46TRF74AB/events.json","paper":"https://pith.science/paper/DEEMQ3QR"},"agent_actions":{"view_html":"https://pith.science/pith/DEEMQ3QRH6D7KQSWY46TRF74AB","download_json":"https://pith.science/pith/DEEMQ3QRH6D7KQSWY46TRF74AB.json","view_paper":"https://pith.science/paper/DEEMQ3QR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.16184&json=true","fetch_graph":"https://pith.science/api/pith-number/DEEMQ3QRH6D7KQSWY46TRF74AB/graph.json","fetch_events":"https://pith.science/api/pith-number/DEEMQ3QRH6D7KQSWY46TRF74AB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DEEMQ3QRH6D7KQSWY46TRF74AB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DEEMQ3QRH6D7KQSWY46TRF74AB/action/storage_attestation","attest_author":"https://pith.science/pith/DEEMQ3QRH6D7KQSWY46TRF74AB/action/author_attestation","sign_citation":"https://pith.science/pith/DEEMQ3QRH6D7KQSWY46TRF74AB/action/citation_signature","submit_replication":"https://pith.science/pith/DEEMQ3QRH6D7KQSWY46TRF74AB/action/replication_record"}},"created_at":"2026-07-05T09:23:31.001120+00:00","updated_at":"2026-07-05T09:23:31.001120+00:00"}