{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:HL5YFUJSVDXGKH6F26QUUOY6MM","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"6364a0b1bb21fce200f2844dc3e6cbe266252a11969831d695fe738dc87a1b9e","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-08T15:38:25Z","title_canon_sha256":"073f5a39139677eda19602a08a374341a7353f3683382dc0490c7a34784f78c8"},"schema_version":"1.0","source":{"id":"2504.06141","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.06141","created_at":"2026-07-05T10:47:57Z"},{"alias_kind":"arxiv_version","alias_value":"2504.06141v2","created_at":"2026-07-05T10:47:57Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.06141","created_at":"2026-07-05T10:47:57Z"},{"alias_kind":"pith_short_12","alias_value":"HL5YFUJSVDXG","created_at":"2026-07-05T10:47:57Z"},{"alias_kind":"pith_short_16","alias_value":"HL5YFUJSVDXGKH6F","created_at":"2026-07-05T10:47:57Z"},{"alias_kind":"pith_short_8","alias_value":"HL5YFUJS","created_at":"2026-07-05T10:47:57Z"}],"graph_snapshots":[{"event_id":"sha256:8c74f9284ae2f0b99770abab4379e09fe607ee71169db0abca4e8e771d1590f4","target":"graph","created_at":"2026-07-05T10:47:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2504.06141/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Reward modeling has emerged as a promising approach for the scalable alignment of language models. However, contemporary reward models (RMs) often lack robustness, awarding high rewards to low-quality, out-of-distribution (OOD) samples. This can lead to reward hacking, where policies exploit unintended shortcuts to maximize rewards, undermining alignment. To address this challenge, we introduce Adv-RM, a novel adversarial training framework that automatically identifies adversarial examples -- responses that receive high rewards from the target RM but are OOD and of low quality. By leveraging ","authors_text":"Adithya Renduchintala, Alexander Bukharin, Haifeng Qian, Oleksii Kuchaiev, Olivier Delalleau, Shengyang Sun, Soumye Singhal, Tuo Zhao, Zhilin Wang","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-08T15:38:25Z","title":"Adversarial Training of Reward Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.06141","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:71c0497d44fa7bf74378669c4f4ac4a802b544364f760632384c1bea906d0981","target":"record","created_at":"2026-07-05T10:47:57Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"6364a0b1bb21fce200f2844dc3e6cbe266252a11969831d695fe738dc87a1b9e","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-08T15:38:25Z","title_canon_sha256":"073f5a39139677eda19602a08a374341a7353f3683382dc0490c7a34784f78c8"},"schema_version":"1.0","source":{"id":"2504.06141","kind":"arxiv","version":2}},"canonical_sha256":"3afb82d132a8ee651fc5d7a14a3b1e63114f83879d84de9eccd8cc460dae18d0","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"3afb82d132a8ee651fc5d7a14a3b1e63114f83879d84de9eccd8cc460dae18d0","first_computed_at":"2026-07-05T10:47:57.957153Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T10:47:57.957153Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"u65yaBIpRlRxQLDP6VIymX9WI6rBRcIOKzsAn8krP/MnhuysNYtjyWK25xuG7ar5rpjgfonGpT5ZJOJQbD57BQ==","signature_status":"signed_v1","signed_at":"2026-07-05T10:47:57.957651Z","signed_message":"canonical_sha256_bytes"},"source_id":"2504.06141","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:71c0497d44fa7bf74378669c4f4ac4a802b544364f760632384c1bea906d0981","sha256:8c74f9284ae2f0b99770abab4379e09fe607ee71169db0abca4e8e771d1590f4"],"state_sha256":"9eaca0efb3f20304c5d1dcd6c3e3d4b163d2bf7cc336b54e74795a7901ba04cf"}