{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:BSYNCCRQNIJBQAGZVVJLBVTMUG","short_pith_number":"pith:BSYNCCRQ","schema_version":"1.0","canonical_sha256":"0cb0d10a306a121800d9ad52b0d66ca1a7f5901fa53134ab92a20ac869b76591","source":{"kind":"arxiv","id":"2605.18309","version":1},"attestation_state":"computed","paper":{"title":"Alignment Dynamics in LLM Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Huanran Chen, Yinpeng Dong, Yuhan Huang","submitted_at":"2026-05-18T12:27:12Z","abstract_excerpt":"Although Large Language Models (LLMs) achieve strong alignment through supervised fine-tuning and reinforcement learning from human feedback, the alignment is often fragile under subsequent fine-tuning. Existing explanations either attribute alignment fragility to gradient geometry or characterize it as a distributional shift in model outputs, yet few provide a unified account that bridges parameter-space learning dynamics with function-space alignment behavior during fine-tuning. In this work, we introduce a tractable alignment score and derive its closed-form update during fine-tuning, yield"},"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":"2605.18309","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-18T12:27:12Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"fc91e59deef2fa1d7132408ced444335c2272af3d2fc3bc73ae3b1fa30ddca37","abstract_canon_sha256":"9f05d33b7eb857ef579bb390986d424e23564ed776486222e72acf5704f6a859"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:05:54.435252Z","signature_b64":"op/JgyzlQypOnPLlyg9bttSRU1ibeghJlI0ATPF0+77xdKy6kUT4E0VUr1gA1Kx8YcZMeHMOdK2g3WTTy624AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0cb0d10a306a121800d9ad52b0d66ca1a7f5901fa53134ab92a20ac869b76591","last_reissued_at":"2026-05-20T00:05:54.434568Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:05:54.434568Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Alignment Dynamics in LLM Fine-Tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Huanran Chen, Yinpeng Dong, Yuhan Huang","submitted_at":"2026-05-18T12:27:12Z","abstract_excerpt":"Although Large Language Models (LLMs) achieve strong alignment through supervised fine-tuning and reinforcement learning from human feedback, the alignment is often fragile under subsequent fine-tuning. Existing explanations either attribute alignment fragility to gradient geometry or characterize it as a distributional shift in model outputs, yet few provide a unified account that bridges parameter-space learning dynamics with function-space alignment behavior during fine-tuning. In this work, we introduce a tractable alignment score and derive its closed-form update during fine-tuning, yield"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.18309","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/2605.18309/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-19T23:33:35.201570Z","status":"skipped","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T23:21:58.884780Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"b7e9773113cb33e7956983cc12e5189041689984cda69ad3041475d9bda96ec4"},"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":"2605.18309","created_at":"2026-05-20T00:05:54.434676+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.18309v1","created_at":"2026-05-20T00:05:54.434676+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.18309","created_at":"2026-05-20T00:05:54.434676+00:00"},{"alias_kind":"pith_short_12","alias_value":"BSYNCCRQNIJB","created_at":"2026-05-20T00:05:54.434676+00:00"},{"alias_kind":"pith_short_16","alias_value":"BSYNCCRQNIJBQAGZ","created_at":"2026-05-20T00:05:54.434676+00:00"},{"alias_kind":"pith_short_8","alias_value":"BSYNCCRQ","created_at":"2026-05-20T00:05:54.434676+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/BSYNCCRQNIJBQAGZVVJLBVTMUG","json":"https://pith.science/pith/BSYNCCRQNIJBQAGZVVJLBVTMUG.json","graph_json":"https://pith.science/api/pith-number/BSYNCCRQNIJBQAGZVVJLBVTMUG/graph.json","events_json":"https://pith.science/api/pith-number/BSYNCCRQNIJBQAGZVVJLBVTMUG/events.json","paper":"https://pith.science/paper/BSYNCCRQ"},"agent_actions":{"view_html":"https://pith.science/pith/BSYNCCRQNIJBQAGZVVJLBVTMUG","download_json":"https://pith.science/pith/BSYNCCRQNIJBQAGZVVJLBVTMUG.json","view_paper":"https://pith.science/paper/BSYNCCRQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.18309&json=true","fetch_graph":"https://pith.science/api/pith-number/BSYNCCRQNIJBQAGZVVJLBVTMUG/graph.json","fetch_events":"https://pith.science/api/pith-number/BSYNCCRQNIJBQAGZVVJLBVTMUG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BSYNCCRQNIJBQAGZVVJLBVTMUG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BSYNCCRQNIJBQAGZVVJLBVTMUG/action/storage_attestation","attest_author":"https://pith.science/pith/BSYNCCRQNIJBQAGZVVJLBVTMUG/action/author_attestation","sign_citation":"https://pith.science/pith/BSYNCCRQNIJBQAGZVVJLBVTMUG/action/citation_signature","submit_replication":"https://pith.science/pith/BSYNCCRQNIJBQAGZVVJLBVTMUG/action/replication_record"}},"created_at":"2026-05-20T00:05:54.434676+00:00","updated_at":"2026-05-20T00:05:54.434676+00:00"}