{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WEDH3WNY44OKNI6LV7JED7C3K7","short_pith_number":"pith:WEDH3WNY","schema_version":"1.0","canonical_sha256":"b1067dd9b8e71ca6a3cbafd241fc5b57e502a7077eeef75a30be02204546052d","source":{"kind":"arxiv","id":"2509.25148","version":2},"attestation_state":"computed","paper":{"title":"AAPA: Adversarially Anchored Preference Alignment for Post-Training of Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Faqiang Qian, Kang An, Liangjian Wen, Mengya Gao, Weikun Zhang, Xuhui Zheng, Yichao Wu, Yong Dai, Ziliang Wang","submitted_at":"2025-09-29T17:53:09Z","abstract_excerpt":"Post-training alignment of large language models often combines supervised fine-tuning (SFT) on expert demonstrations with reinforcement learning (RL) from preference or verifiable feedback. SFT provides a useful behavioral anchor but can overfit to static demonstrations, whereas RL encourages exploration but may drift from expert behavior or exploit imperfect rewards. We propose \\textbf{AAPA} (\\emph{Adversarially Anchored Preference Alignment}), a plug-in framework that augments existing post-training objectives with a sentence-level adversarial anchoring signal. AAPA compares policy rollouts"},"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":"2509.25148","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2025-09-29T17:53:09Z","cross_cats_sorted":[],"title_canon_sha256":"5a75cc940a70669a15a011662be42fe98e8dd36c2d7302ac1b57040c2d83c0fa","abstract_canon_sha256":"52e1a106602410e5a820cefad8871ee84ab7c69eb4bb4245082ca069a35b7624"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:12:15.287734Z","signature_b64":"/XW+mHMMe9YTc15cyswRsqatawCg/18LW7s8ZooyoMx6BuJDbmZhBv4/J9KJBqXXasI2Kyvs5YH4yuIVqapdDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b1067dd9b8e71ca6a3cbafd241fc5b57e502a7077eeef75a30be02204546052d","last_reissued_at":"2026-06-19T16:12:15.287284Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:12:15.287284Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AAPA: Adversarially Anchored Preference Alignment for Post-Training of Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Faqiang Qian, Kang An, Liangjian Wen, Mengya Gao, Weikun Zhang, Xuhui Zheng, Yichao Wu, Yong Dai, Ziliang Wang","submitted_at":"2025-09-29T17:53:09Z","abstract_excerpt":"Post-training alignment of large language models often combines supervised fine-tuning (SFT) on expert demonstrations with reinforcement learning (RL) from preference or verifiable feedback. SFT provides a useful behavioral anchor but can overfit to static demonstrations, whereas RL encourages exploration but may drift from expert behavior or exploit imperfect rewards. We propose \\textbf{AAPA} (\\emph{Adversarially Anchored Preference Alignment}), a plug-in framework that augments existing post-training objectives with a sentence-level adversarial anchoring signal. AAPA compares policy rollouts"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.25148","kind":"arxiv","version":2},"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/2509.25148/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":"2509.25148","created_at":"2026-06-19T16:12:15.287345+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.25148v2","created_at":"2026-06-19T16:12:15.287345+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.25148","created_at":"2026-06-19T16:12:15.287345+00:00"},{"alias_kind":"pith_short_12","alias_value":"WEDH3WNY44OK","created_at":"2026-06-19T16:12:15.287345+00:00"},{"alias_kind":"pith_short_16","alias_value":"WEDH3WNY44OKNI6L","created_at":"2026-06-19T16:12:15.287345+00:00"},{"alias_kind":"pith_short_8","alias_value":"WEDH3WNY","created_at":"2026-06-19T16:12:15.287345+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/WEDH3WNY44OKNI6LV7JED7C3K7","json":"https://pith.science/pith/WEDH3WNY44OKNI6LV7JED7C3K7.json","graph_json":"https://pith.science/api/pith-number/WEDH3WNY44OKNI6LV7JED7C3K7/graph.json","events_json":"https://pith.science/api/pith-number/WEDH3WNY44OKNI6LV7JED7C3K7/events.json","paper":"https://pith.science/paper/WEDH3WNY"},"agent_actions":{"view_html":"https://pith.science/pith/WEDH3WNY44OKNI6LV7JED7C3K7","download_json":"https://pith.science/pith/WEDH3WNY44OKNI6LV7JED7C3K7.json","view_paper":"https://pith.science/paper/WEDH3WNY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.25148&json=true","fetch_graph":"https://pith.science/api/pith-number/WEDH3WNY44OKNI6LV7JED7C3K7/graph.json","fetch_events":"https://pith.science/api/pith-number/WEDH3WNY44OKNI6LV7JED7C3K7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WEDH3WNY44OKNI6LV7JED7C3K7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WEDH3WNY44OKNI6LV7JED7C3K7/action/storage_attestation","attest_author":"https://pith.science/pith/WEDH3WNY44OKNI6LV7JED7C3K7/action/author_attestation","sign_citation":"https://pith.science/pith/WEDH3WNY44OKNI6LV7JED7C3K7/action/citation_signature","submit_replication":"https://pith.science/pith/WEDH3WNY44OKNI6LV7JED7C3K7/action/replication_record"}},"created_at":"2026-06-19T16:12:15.287345+00:00","updated_at":"2026-06-19T16:12:15.287345+00:00"}