{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SAD3VLBWBP2QX2Z6F5O3JEXDEI","short_pith_number":"pith:SAD3VLBW","schema_version":"1.0","canonical_sha256":"9007baac360bf50beb3e2f5db492e322302febdf125e5db9c3c44a55b46329b2","source":{"kind":"arxiv","id":"2605.25629","version":1},"attestation_state":"computed","paper":{"title":"When In-Distribution Gains Fail: Evaluating Weak-to-Strong Reward Models under Preference Shift","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Anh Tuan Luu, Cong-Duy Nguyen, Khoi Le, Miao Chunyan, Phong Nguyen, See-kiong Ng, Thong Nguyen, Tri Cao","submitted_at":"2026-05-25T09:30:49Z","abstract_excerpt":"Weak-to-strong (W2S) generalization is a promising framework for scalable oversight, yet existing evaluations often test students under matched train--test distributions. Therefore, we study W2S preference learning under zero-shot distribution shift and find that strong students trained on weak preference labels can appear successful in-distribution while failing to transfer across preference datasets. We provide evidence for a representational failure mode in which weak-supervised fine-tuning can pull the strong model toward source-domain features instead of maintaining broadly transferable p"},"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":"2605.25629","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-05-25T09:30:49Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"7303dda99b4f2611f3bbc2a7ffe49fac1d14734dbf1b716bdf61961d9af878f0","abstract_canon_sha256":"aee6572a9d7cffc2dd11ba39cfb9ee20a9168bb047b7cb6e14bf266fb6af11ae"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:47.276420Z","signature_b64":"V0YB4jUx44l5xFT01NA+um1MPgfxuy45Dg37evCWZg/i9pBa3vtduEGpxWyhSCXUyCdBKQWqNHsI8wYUrVObDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9007baac360bf50beb3e2f5db492e322302febdf125e5db9c3c44a55b46329b2","last_reissued_at":"2026-05-26T02:04:47.275821Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:47.275821Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When In-Distribution Gains Fail: Evaluating Weak-to-Strong Reward Models under Preference Shift","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Anh Tuan Luu, Cong-Duy Nguyen, Khoi Le, Miao Chunyan, Phong Nguyen, See-kiong Ng, Thong Nguyen, Tri Cao","submitted_at":"2026-05-25T09:30:49Z","abstract_excerpt":"Weak-to-strong (W2S) generalization is a promising framework for scalable oversight, yet existing evaluations often test students under matched train--test distributions. Therefore, we study W2S preference learning under zero-shot distribution shift and find that strong students trained on weak preference labels can appear successful in-distribution while failing to transfer across preference datasets. We provide evidence for a representational failure mode in which weak-supervised fine-tuning can pull the strong model toward source-domain features instead of maintaining broadly transferable p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.25629","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/2605.25629/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":"2605.25629","created_at":"2026-05-26T02:04:47.275916+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.25629v1","created_at":"2026-05-26T02:04:47.275916+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.25629","created_at":"2026-05-26T02:04:47.275916+00:00"},{"alias_kind":"pith_short_12","alias_value":"SAD3VLBWBP2Q","created_at":"2026-05-26T02:04:47.275916+00:00"},{"alias_kind":"pith_short_16","alias_value":"SAD3VLBWBP2QX2Z6","created_at":"2026-05-26T02:04:47.275916+00:00"},{"alias_kind":"pith_short_8","alias_value":"SAD3VLBW","created_at":"2026-05-26T02:04:47.275916+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/SAD3VLBWBP2QX2Z6F5O3JEXDEI","json":"https://pith.science/pith/SAD3VLBWBP2QX2Z6F5O3JEXDEI.json","graph_json":"https://pith.science/api/pith-number/SAD3VLBWBP2QX2Z6F5O3JEXDEI/graph.json","events_json":"https://pith.science/api/pith-number/SAD3VLBWBP2QX2Z6F5O3JEXDEI/events.json","paper":"https://pith.science/paper/SAD3VLBW"},"agent_actions":{"view_html":"https://pith.science/pith/SAD3VLBWBP2QX2Z6F5O3JEXDEI","download_json":"https://pith.science/pith/SAD3VLBWBP2QX2Z6F5O3JEXDEI.json","view_paper":"https://pith.science/paper/SAD3VLBW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.25629&json=true","fetch_graph":"https://pith.science/api/pith-number/SAD3VLBWBP2QX2Z6F5O3JEXDEI/graph.json","fetch_events":"https://pith.science/api/pith-number/SAD3VLBWBP2QX2Z6F5O3JEXDEI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SAD3VLBWBP2QX2Z6F5O3JEXDEI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SAD3VLBWBP2QX2Z6F5O3JEXDEI/action/storage_attestation","attest_author":"https://pith.science/pith/SAD3VLBWBP2QX2Z6F5O3JEXDEI/action/author_attestation","sign_citation":"https://pith.science/pith/SAD3VLBWBP2QX2Z6F5O3JEXDEI/action/citation_signature","submit_replication":"https://pith.science/pith/SAD3VLBWBP2QX2Z6F5O3JEXDEI/action/replication_record"}},"created_at":"2026-05-26T02:04:47.275916+00:00","updated_at":"2026-05-26T02:04:47.275916+00:00"}