{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NASLEKJR6XI6TZEGFZ23I53JVV","short_pith_number":"pith:NASLEKJR","schema_version":"1.0","canonical_sha256":"6824b22931f5d1e9e4862e75b47769ad6594e0b276c76cbab318388e02e10155","source":{"kind":"arxiv","id":"2606.09866","version":1},"attestation_state":"computed","paper":{"title":"Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Di Gao, Jianhao Zhang, Ou Wu, Xinrui Chen","submitted_at":"2026-06-01T02:43:36Z","abstract_excerpt":"Fine-tuning safety aligned large language models (LLMs) on downstream data improves adaptation but may erode learned safety behavior. Existing methods use fixed safety examples, global constraints, or one-sided task filtering. Our diagnostics show task updates expose different safety constraints, motivating joint selection of relevant references and compatible task samples. We propose DualSelect, a coupled framework for task and reference selection that refreshes task conditioned safety references before filtering whole task samples compatible with the induced reference direction. Under a mini"},"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":"2606.09866","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-01T02:43:36Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"cf3b5402cfdc882febd3bba21d3990797400fc292b6647271a7131bbdfaedc86","abstract_canon_sha256":"ad6a1ce5a9104b17a29b7738b2b5c8c57a311bf526e9f002ee7927f51381a21f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-10T00:08:30.278308Z","signature_b64":"L03kz9M62e3RkDJbk6PzqjarLMo1mC7nbMqIhJdMTg3Hf+Ic+pdBi0RhSFzJzNVKwAVTQRWryyBcRBYFPH0yBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6824b22931f5d1e9e4862e75b47769ad6594e0b276c76cbab318388e02e10155","last_reissued_at":"2026-06-10T00:08:30.277489Z","signature_status":"signed_v1","first_computed_at":"2026-06-10T00:08:30.277489Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Two to Tango: Coupled Task-Reference Selection for Safe LLM Fine-tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Di Gao, Jianhao Zhang, Ou Wu, Xinrui Chen","submitted_at":"2026-06-01T02:43:36Z","abstract_excerpt":"Fine-tuning safety aligned large language models (LLMs) on downstream data improves adaptation but may erode learned safety behavior. Existing methods use fixed safety examples, global constraints, or one-sided task filtering. Our diagnostics show task updates expose different safety constraints, motivating joint selection of relevant references and compatible task samples. We propose DualSelect, a coupled framework for task and reference selection that refreshes task conditioned safety references before filtering whole task samples compatible with the induced reference direction. Under a mini"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.09866","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/2606.09866/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":"2606.09866","created_at":"2026-06-10T00:08:30.277637+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.09866v1","created_at":"2026-06-10T00:08:30.277637+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.09866","created_at":"2026-06-10T00:08:30.277637+00:00"},{"alias_kind":"pith_short_12","alias_value":"NASLEKJR6XI6","created_at":"2026-06-10T00:08:30.277637+00:00"},{"alias_kind":"pith_short_16","alias_value":"NASLEKJR6XI6TZEG","created_at":"2026-06-10T00:08:30.277637+00:00"},{"alias_kind":"pith_short_8","alias_value":"NASLEKJR","created_at":"2026-06-10T00:08:30.277637+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/NASLEKJR6XI6TZEGFZ23I53JVV","json":"https://pith.science/pith/NASLEKJR6XI6TZEGFZ23I53JVV.json","graph_json":"https://pith.science/api/pith-number/NASLEKJR6XI6TZEGFZ23I53JVV/graph.json","events_json":"https://pith.science/api/pith-number/NASLEKJR6XI6TZEGFZ23I53JVV/events.json","paper":"https://pith.science/paper/NASLEKJR"},"agent_actions":{"view_html":"https://pith.science/pith/NASLEKJR6XI6TZEGFZ23I53JVV","download_json":"https://pith.science/pith/NASLEKJR6XI6TZEGFZ23I53JVV.json","view_paper":"https://pith.science/paper/NASLEKJR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.09866&json=true","fetch_graph":"https://pith.science/api/pith-number/NASLEKJR6XI6TZEGFZ23I53JVV/graph.json","fetch_events":"https://pith.science/api/pith-number/NASLEKJR6XI6TZEGFZ23I53JVV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NASLEKJR6XI6TZEGFZ23I53JVV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NASLEKJR6XI6TZEGFZ23I53JVV/action/storage_attestation","attest_author":"https://pith.science/pith/NASLEKJR6XI6TZEGFZ23I53JVV/action/author_attestation","sign_citation":"https://pith.science/pith/NASLEKJR6XI6TZEGFZ23I53JVV/action/citation_signature","submit_replication":"https://pith.science/pith/NASLEKJR6XI6TZEGFZ23I53JVV/action/replication_record"}},"created_at":"2026-06-10T00:08:30.277637+00:00","updated_at":"2026-06-10T00:08:30.277637+00:00"}