{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:PVG73FUFVM2W5O5TAGTS457QF3","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":"7cf849e07bfa31690c04496f70cfdcc5ef97ce2c7becb29977e80196f161735f","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-11T18:41:12Z","title_canon_sha256":"64f159179108eb2e6201f854d383fb9a0adf09c38557b327780868ad7c36db05"},"schema_version":"1.0","source":{"id":"2605.11134","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.11134","created_at":"2026-06-01T02:03:43Z"},{"alias_kind":"arxiv_version","alias_value":"2605.11134v2","created_at":"2026-06-01T02:03:43Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.11134","created_at":"2026-06-01T02:03:43Z"},{"alias_kind":"pith_short_12","alias_value":"PVG73FUFVM2W","created_at":"2026-06-01T02:03:43Z"},{"alias_kind":"pith_short_16","alias_value":"PVG73FUFVM2W5O5T","created_at":"2026-06-01T02:03:43Z"},{"alias_kind":"pith_short_8","alias_value":"PVG73FUF","created_at":"2026-06-01T02:03:43Z"}],"graph_snapshots":[{"event_id":"sha256:491abb27088bfa4b4f8461d66a438028b4a54b0134539bb0629c5c79945de798","target":"graph","created_at":"2026-06-01T02:03:43Z","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":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"standard preference-learning objectives induce reliance on spurious features at the population level through two channels: mean spurious bias and causal--spurious correlation leakage. We then show that this reliance creates an irreducible vulnerability to distribution shift: more data from the same training distribution fails to reduce the model's dependence on spurious features."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The mechanisms and mitigation identified for log-linear policies extend to neural networks and large language models without significant degradation of causal learning."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Standard preference learning induces spurious feature reliance via mean bias and correlation leakage, creating irreducible distribution shift vulnerabilities that tie training mitigates without degrading causal learning."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Preference optimization induces spurious feature reliance through mean bias and correlation leakage, creating a vulnerability to distribution shift that more training data cannot fix."}],"snapshot_sha256":"3d4c018c428be0efcf7712681d0b46f0f5156ef73311d8d5521937ca812ed657"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T05:02:00.664652Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T13:34:46.818079Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T10:31:16.736109Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T08:45:17.420638Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.11134/integrity.json","findings":[],"snapshot_sha256":"979b59de78302d016be40cece20f3b48bbccd9ae085f6175d2bb9a54f70207b5","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Preference learning methods like Direct Preference Optimization (DPO) are known to induce reliance on spurious correlations, leading to sycophancy and length bias in today's language models and potentially severe goal misgeneralization in future systems. In this work, we provide a unified theoretical analysis of this phenomenon, characterizing the mechanisms of spurious learning, its consequences on deployment, and a provable mitigation strategy. Focusing on log-linear policies, we show that standard preference-learning objectives induce reliance on spurious features at the population level th","authors_text":"Alex Semendinger, Christian Moya, Elliott Thornley, Guang Lin","cross_cats":["cs.AI"],"headline":"Preference optimization induces spurious feature reliance through mean bias and correlation leakage, creating a vulnerability to distribution shift that more training data cannot fix.","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-11T18:41:12Z","title":"Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.11134","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-13T06:26:36.426012Z","id":"623f885e-78f1-4599-bc85-e2dbfa0f12c8","model_set":{"reader":"grok-4.3"},"one_line_summary":"Standard preference learning induces spurious feature reliance via mean bias and correlation leakage, creating irreducible distribution shift vulnerabilities that tie training mitigates without degrading causal learning.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Preference optimization induces spurious feature reliance through mean bias and correlation leakage, creating a vulnerability to distribution shift that more training data cannot fix.","strongest_claim":"standard preference-learning objectives induce reliance on spurious features at the population level through two channels: mean spurious bias and causal--spurious correlation leakage. We then show that this reliance creates an irreducible vulnerability to distribution shift: more data from the same training distribution fails to reduce the model's dependence on spurious features.","weakest_assumption":"The mechanisms and mitigation identified for log-linear policies extend to neural networks and large language models without significant degradation of causal learning."}},"verdict_id":"623f885e-78f1-4599-bc85-e2dbfa0f12c8"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:0c39653bbf65888dded8b188ba70bdb732f524c704fb9600edfd8b7d1beb6841","target":"record","created_at":"2026-06-01T02:03:43Z","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":"7cf849e07bfa31690c04496f70cfdcc5ef97ce2c7becb29977e80196f161735f","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-11T18:41:12Z","title_canon_sha256":"64f159179108eb2e6201f854d383fb9a0adf09c38557b327780868ad7c36db05"},"schema_version":"1.0","source":{"id":"2605.11134","kind":"arxiv","version":2}},"canonical_sha256":"7d4dfd9685ab356ebbb301a72e77f02efc669222652cd3613718eb87310fec74","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"7d4dfd9685ab356ebbb301a72e77f02efc669222652cd3613718eb87310fec74","first_computed_at":"2026-06-01T02:03:43.357095Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-01T02:03:43.357095Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"5suPHe08zPBL7FPvTcr5QNPIjCbH6bLNu7Rw2ReYxYR6u03TOo2ZY1lI5UspWkGSfhP8TVhg8MwHzXQe2VxBCw==","signature_status":"signed_v1","signed_at":"2026-06-01T02:03:43.358497Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.11134","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:0c39653bbf65888dded8b188ba70bdb732f524c704fb9600edfd8b7d1beb6841","sha256:491abb27088bfa4b4f8461d66a438028b4a54b0134539bb0629c5c79945de798"],"state_sha256":"5bb405e6a6ae9b7144742999301d57d3576c8bdb7d5eb68a3cd2d3a4afe0a3c7"}