{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:5HKMIHUSFDMFIJ67MBVQJFNPSQ","short_pith_number":"pith:5HKMIHUS","schema_version":"1.0","canonical_sha256":"e9d4c41e9228d85427df606b0495af942d76a3bfaac365c654ae481899420a15","source":{"kind":"arxiv","id":"2507.15906","version":1},"attestation_state":"computed","paper":{"title":"Towards Reliable, Uncertainty-Aware Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aditya Gopalan, Debangshu Banerjee, Kintan Saha","submitted_at":"2025-07-21T12:51:29Z","abstract_excerpt":"Alignment of large language models (LLMs) typically involves training a reward model on preference data, followed by policy optimization with respect to the reward model. However, optimizing policies with respect to a single reward model estimate can render it vulnerable to inaccuracies in the reward model. We empirically study the variability of reward model training on open-source benchmarks. We observe that independently trained reward models on the same preference dataset can exhibit substantial disagreement, highlighting the instability of current alignment strategies. Employing a theoret"},"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":"2507.15906","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-21T12:51:29Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"6f9cae5b8982a82d0d8c18d75b37cfb1bb20275c1ffef9277c69cd2ab6e8939d","abstract_canon_sha256":"eb75b926407cb9e1eb3032ba6bd5488748401700c0d2f472b8a6106a5e2ffbc4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:40:59.047709Z","signature_b64":"6iQvxCY+y5kcy7sRMunKIRSRbe3C5yq2L0PT75QKdrZlHRV1RihWFcSsqbLophkXMbirZtiQ6KbcbC2kXS1ZDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e9d4c41e9228d85427df606b0495af942d76a3bfaac365c654ae481899420a15","last_reissued_at":"2026-07-05T11:40:59.047094Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:40:59.047094Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Reliable, Uncertainty-Aware Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Aditya Gopalan, Debangshu Banerjee, Kintan Saha","submitted_at":"2025-07-21T12:51:29Z","abstract_excerpt":"Alignment of large language models (LLMs) typically involves training a reward model on preference data, followed by policy optimization with respect to the reward model. However, optimizing policies with respect to a single reward model estimate can render it vulnerable to inaccuracies in the reward model. We empirically study the variability of reward model training on open-source benchmarks. We observe that independently trained reward models on the same preference dataset can exhibit substantial disagreement, highlighting the instability of current alignment strategies. Employing a theoret"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.15906","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/2507.15906/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":"2507.15906","created_at":"2026-07-05T11:40:59.047160+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.15906v1","created_at":"2026-07-05T11:40:59.047160+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.15906","created_at":"2026-07-05T11:40:59.047160+00:00"},{"alias_kind":"pith_short_12","alias_value":"5HKMIHUSFDMF","created_at":"2026-07-05T11:40:59.047160+00:00"},{"alias_kind":"pith_short_16","alias_value":"5HKMIHUSFDMFIJ67","created_at":"2026-07-05T11:40:59.047160+00:00"},{"alias_kind":"pith_short_8","alias_value":"5HKMIHUS","created_at":"2026-07-05T11:40:59.047160+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5HKMIHUSFDMFIJ67MBVQJFNPSQ","json":"https://pith.science/pith/5HKMIHUSFDMFIJ67MBVQJFNPSQ.json","graph_json":"https://pith.science/api/pith-number/5HKMIHUSFDMFIJ67MBVQJFNPSQ/graph.json","events_json":"https://pith.science/api/pith-number/5HKMIHUSFDMFIJ67MBVQJFNPSQ/events.json","paper":"https://pith.science/paper/5HKMIHUS"},"agent_actions":{"view_html":"https://pith.science/pith/5HKMIHUSFDMFIJ67MBVQJFNPSQ","download_json":"https://pith.science/pith/5HKMIHUSFDMFIJ67MBVQJFNPSQ.json","view_paper":"https://pith.science/paper/5HKMIHUS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.15906&json=true","fetch_graph":"https://pith.science/api/pith-number/5HKMIHUSFDMFIJ67MBVQJFNPSQ/graph.json","fetch_events":"https://pith.science/api/pith-number/5HKMIHUSFDMFIJ67MBVQJFNPSQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5HKMIHUSFDMFIJ67MBVQJFNPSQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5HKMIHUSFDMFIJ67MBVQJFNPSQ/action/storage_attestation","attest_author":"https://pith.science/pith/5HKMIHUSFDMFIJ67MBVQJFNPSQ/action/author_attestation","sign_citation":"https://pith.science/pith/5HKMIHUSFDMFIJ67MBVQJFNPSQ/action/citation_signature","submit_replication":"https://pith.science/pith/5HKMIHUSFDMFIJ67MBVQJFNPSQ/action/replication_record"}},"created_at":"2026-07-05T11:40:59.047160+00:00","updated_at":"2026-07-05T11:40:59.047160+00:00"}