{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2023:3XGKVZ7N2X3QZMTXCD3SYGUBWT","short_pith_number":"pith:3XGKVZ7N","canonical_record":{"source":{"id":"2308.13449","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-08-25T15:51:15Z","cross_cats_sorted":[],"title_canon_sha256":"3270cb29772400247e89960336f93fb593135f6540b8948c998d8c9ddcc03490","abstract_canon_sha256":"3727b343e63e752f64d891f0f76f9a624ea0c73451bcd3f610ff51e4555ab629"},"schema_version":"1.0"},"canonical_sha256":"ddccaae7edd5f70cb27710f72c1a81b4e7eb902bfaaf20318c5c3accfc78b504","source":{"kind":"arxiv","id":"2308.13449","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2308.13449","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"arxiv_version","alias_value":"2308.13449v1","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2308.13449","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"pith_short_12","alias_value":"3XGKVZ7N2X3Q","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"pith_short_16","alias_value":"3XGKVZ7N2X3QZMTX","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"pith_short_8","alias_value":"3XGKVZ7N","created_at":"2026-07-05T06:44:44Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2023:3XGKVZ7N2X3QZMTXCD3SYGUBWT","target":"record","payload":{"canonical_record":{"source":{"id":"2308.13449","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-08-25T15:51:15Z","cross_cats_sorted":[],"title_canon_sha256":"3270cb29772400247e89960336f93fb593135f6540b8948c998d8c9ddcc03490","abstract_canon_sha256":"3727b343e63e752f64d891f0f76f9a624ea0c73451bcd3f610ff51e4555ab629"},"schema_version":"1.0"},"canonical_sha256":"ddccaae7edd5f70cb27710f72c1a81b4e7eb902bfaaf20318c5c3accfc78b504","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:44:44.154376Z","signature_b64":"THeXMiaR3Pkc7MvpKUcPwisA6WVNjKXGOoEUVinWYJ4nt1PuNWafkL9iKsmwgE7XaDJSsQ43x5MckeQZiLj3DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ddccaae7edd5f70cb27710f72c1a81b4e7eb902bfaaf20318c5c3accfc78b504","last_reissued_at":"2026-07-05T06:44:44.153801Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:44:44.153801Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2308.13449","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T06:44:44Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"1vCqelin6rFqV3W0h3vDniZj+HCeI1ss/5q0vGPk7sBen4rrBuA13EG3L7kHkVHZhT+t94NWR6VZ8lGi5Y5fAQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T17:55:55.382717Z"},"content_sha256":"cd66a8590e50cfc40f15677717e0d5860cbf43199bedb95c47591e98a4feb297","schema_version":"1.0","event_id":"sha256:cd66a8590e50cfc40f15677717e0d5860cbf43199bedb95c47591e98a4feb297"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2023:3XGKVZ7N2X3QZMTXCD3SYGUBWT","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"The Poison of Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Aibek Bekbayev, James Yamazaki, Sungbae Chun, Yerzat Dulat","submitted_at":"2023-08-25T15:51:15Z","abstract_excerpt":"From the perspective of content safety issues, alignment has shown to limit large language models' (LLMs) harmful content generation. This intentional method of reinforcing models to not respond to certain user inputs seem to be present in many modern open-source instruction tuning datasets such as OpenAssistant or Guanaco. We introduce a novel insight to an instruction-tuned model's performance affected by the presence of alignment in supervised fine-tuning dataset. To be specific, we noticed that alignment acts as if it is poisoning the instruction dataset. Experimentally, we demonstrate tha"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2308.13449","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/2308.13449/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T06:44:44Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"VW2cKBUX0RcVvgQ7joOgUWpeo6DXG6LH2ocbDrbcpZO4W2WPCc0HvTnuOYu44uMj97BgxSlRDk7WWgkKnwJTAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-08T17:55:55.383662Z"},"content_sha256":"455f0feaea005312eb0cea8a1f07773ea457ef3d4d21a3408a55363221888b51","schema_version":"1.0","event_id":"sha256:455f0feaea005312eb0cea8a1f07773ea457ef3d4d21a3408a55363221888b51"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/3XGKVZ7N2X3QZMTXCD3SYGUBWT/bundle.json","state_url":"https://pith.science/pith/3XGKVZ7N2X3QZMTXCD3SYGUBWT/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/3XGKVZ7N2X3QZMTXCD3SYGUBWT/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-08T17:55:55Z","links":{"resolver":"https://pith.science/pith/3XGKVZ7N2X3QZMTXCD3SYGUBWT","bundle":"https://pith.science/pith/3XGKVZ7N2X3QZMTXCD3SYGUBWT/bundle.json","state":"https://pith.science/pith/3XGKVZ7N2X3QZMTXCD3SYGUBWT/state.json","well_known_bundle":"https://pith.science/.well-known/pith/3XGKVZ7N2X3QZMTXCD3SYGUBWT/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2023:3XGKVZ7N2X3QZMTXCD3SYGUBWT","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":"3727b343e63e752f64d891f0f76f9a624ea0c73451bcd3f610ff51e4555ab629","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-08-25T15:51:15Z","title_canon_sha256":"3270cb29772400247e89960336f93fb593135f6540b8948c998d8c9ddcc03490"},"schema_version":"1.0","source":{"id":"2308.13449","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2308.13449","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"arxiv_version","alias_value":"2308.13449v1","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2308.13449","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"pith_short_12","alias_value":"3XGKVZ7N2X3Q","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"pith_short_16","alias_value":"3XGKVZ7N2X3QZMTX","created_at":"2026-07-05T06:44:44Z"},{"alias_kind":"pith_short_8","alias_value":"3XGKVZ7N","created_at":"2026-07-05T06:44:44Z"}],"graph_snapshots":[{"event_id":"sha256:455f0feaea005312eb0cea8a1f07773ea457ef3d4d21a3408a55363221888b51","target":"graph","created_at":"2026-07-05T06:44:44Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2308.13449/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"From the perspective of content safety issues, alignment has shown to limit large language models' (LLMs) harmful content generation. This intentional method of reinforcing models to not respond to certain user inputs seem to be present in many modern open-source instruction tuning datasets such as OpenAssistant or Guanaco. We introduce a novel insight to an instruction-tuned model's performance affected by the presence of alignment in supervised fine-tuning dataset. To be specific, we noticed that alignment acts as if it is poisoning the instruction dataset. Experimentally, we demonstrate tha","authors_text":"Aibek Bekbayev, James Yamazaki, Sungbae Chun, Yerzat Dulat","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-08-25T15:51:15Z","title":"The Poison of Alignment"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2308.13449","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:cd66a8590e50cfc40f15677717e0d5860cbf43199bedb95c47591e98a4feb297","target":"record","created_at":"2026-07-05T06:44:44Z","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":"3727b343e63e752f64d891f0f76f9a624ea0c73451bcd3f610ff51e4555ab629","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-08-25T15:51:15Z","title_canon_sha256":"3270cb29772400247e89960336f93fb593135f6540b8948c998d8c9ddcc03490"},"schema_version":"1.0","source":{"id":"2308.13449","kind":"arxiv","version":1}},"canonical_sha256":"ddccaae7edd5f70cb27710f72c1a81b4e7eb902bfaaf20318c5c3accfc78b504","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ddccaae7edd5f70cb27710f72c1a81b4e7eb902bfaaf20318c5c3accfc78b504","first_computed_at":"2026-07-05T06:44:44.153801Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T06:44:44.153801Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"THeXMiaR3Pkc7MvpKUcPwisA6WVNjKXGOoEUVinWYJ4nt1PuNWafkL9iKsmwgE7XaDJSsQ43x5MckeQZiLj3DA==","signature_status":"signed_v1","signed_at":"2026-07-05T06:44:44.154376Z","signed_message":"canonical_sha256_bytes"},"source_id":"2308.13449","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:cd66a8590e50cfc40f15677717e0d5860cbf43199bedb95c47591e98a4feb297","sha256:455f0feaea005312eb0cea8a1f07773ea457ef3d4d21a3408a55363221888b51"],"state_sha256":"c8b429d4a040471c8319ad18f18f78c02d58df72c3c0dbe5e79bebc96646b1b3"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"aqTMrk53Tvm1uAUPyPV6CCJb2cp/EgMVRyPehLctJibXJqDiq2xHu6xpLKSPwo7Je7m7VOgjV3Z6smVDVomXDg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-08T17:55:55.389605Z","bundle_sha256":"63f699831d59b1cbb796715902e93c0f9f0b980ec575e64b76283a9084044fb2"}}