{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CIDCZ3O6H3RG4WWCPPTSB6S2KV","short_pith_number":"pith:CIDCZ3O6","schema_version":"1.0","canonical_sha256":"12062cedde3ee26e5ac27be720fa5a5551d3f02d9ce7ce7e8f932feee96a6f04","source":{"kind":"arxiv","id":"2509.26169","version":2},"attestation_state":"computed","paper":{"title":"Alignment-Aware Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Fr\\'ed\\'eric Berdoz, Luca A. Lanzend\\\"orfer, Ren\\'e Caky, Roger Wattenhofer","submitted_at":"2025-09-30T12:24:43Z","abstract_excerpt":"Alignment of large language models remains a central challenge in natural language processing. Preference optimization has emerged as a popular and effective method for improving alignment, typically through training-time or prompt-based interventions. In this paper, we introduce alignment-aware decoding (AAD), a method to enhance model alignment directly at inference. Theoretically, AAD can be interpreted as implicit reward optimization, yet it requires no specialized training beyond the standard DPO setup. Empirically, AAD consistently outperforms strong baselines across diverse alignment be"},"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":"2509.26169","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-09-30T12:24:43Z","cross_cats_sorted":[],"title_canon_sha256":"b6f496d48bfb239a2f5ea7de9a7886242be813879cb2d9dd45998402e03f8cdb","abstract_canon_sha256":"5a6f0c415a720d181ed06577845ee05f66290f0024692daedac59963d72b514b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-03T02:05:42.519023Z","signature_b64":"xHIoseOZFo1eLTbHU4/bnw0cLaYSvFPA1AZA+I27XL0oS5ugx6EEyIzT8H6dbO8JWhESRsECPQbYsGneT7yPCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"12062cedde3ee26e5ac27be720fa5a5551d3f02d9ce7ce7e8f932feee96a6f04","last_reissued_at":"2026-06-03T02:05:42.518529Z","signature_status":"signed_v1","first_computed_at":"2026-06-03T02:05:42.518529Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Alignment-Aware Decoding","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Fr\\'ed\\'eric Berdoz, Luca A. Lanzend\\\"orfer, Ren\\'e Caky, Roger Wattenhofer","submitted_at":"2025-09-30T12:24:43Z","abstract_excerpt":"Alignment of large language models remains a central challenge in natural language processing. Preference optimization has emerged as a popular and effective method for improving alignment, typically through training-time or prompt-based interventions. In this paper, we introduce alignment-aware decoding (AAD), a method to enhance model alignment directly at inference. Theoretically, AAD can be interpreted as implicit reward optimization, yet it requires no specialized training beyond the standard DPO setup. Empirically, AAD consistently outperforms strong baselines across diverse alignment be"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.26169","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/2509.26169/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":"2509.26169","created_at":"2026-06-03T02:05:42.518596+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.26169v2","created_at":"2026-06-03T02:05:42.518596+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.26169","created_at":"2026-06-03T02:05:42.518596+00:00"},{"alias_kind":"pith_short_12","alias_value":"CIDCZ3O6H3RG","created_at":"2026-06-03T02:05:42.518596+00:00"},{"alias_kind":"pith_short_16","alias_value":"CIDCZ3O6H3RG4WWC","created_at":"2026-06-03T02:05:42.518596+00:00"},{"alias_kind":"pith_short_8","alias_value":"CIDCZ3O6","created_at":"2026-06-03T02:05:42.518596+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.23244","citing_title":"Convex Optimization for Alignment and Preference Learning on a Single GPU","ref_index":13,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CIDCZ3O6H3RG4WWCPPTSB6S2KV","json":"https://pith.science/pith/CIDCZ3O6H3RG4WWCPPTSB6S2KV.json","graph_json":"https://pith.science/api/pith-number/CIDCZ3O6H3RG4WWCPPTSB6S2KV/graph.json","events_json":"https://pith.science/api/pith-number/CIDCZ3O6H3RG4WWCPPTSB6S2KV/events.json","paper":"https://pith.science/paper/CIDCZ3O6"},"agent_actions":{"view_html":"https://pith.science/pith/CIDCZ3O6H3RG4WWCPPTSB6S2KV","download_json":"https://pith.science/pith/CIDCZ3O6H3RG4WWCPPTSB6S2KV.json","view_paper":"https://pith.science/paper/CIDCZ3O6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.26169&json=true","fetch_graph":"https://pith.science/api/pith-number/CIDCZ3O6H3RG4WWCPPTSB6S2KV/graph.json","fetch_events":"https://pith.science/api/pith-number/CIDCZ3O6H3RG4WWCPPTSB6S2KV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CIDCZ3O6H3RG4WWCPPTSB6S2KV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CIDCZ3O6H3RG4WWCPPTSB6S2KV/action/storage_attestation","attest_author":"https://pith.science/pith/CIDCZ3O6H3RG4WWCPPTSB6S2KV/action/author_attestation","sign_citation":"https://pith.science/pith/CIDCZ3O6H3RG4WWCPPTSB6S2KV/action/citation_signature","submit_replication":"https://pith.science/pith/CIDCZ3O6H3RG4WWCPPTSB6S2KV/action/replication_record"}},"created_at":"2026-06-03T02:05:42.518596+00:00","updated_at":"2026-06-03T02:05:42.518596+00:00"}