{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:VDQNNFDM7HH726DYZD45TQ7B3Z","short_pith_number":"pith:VDQNNFDM","canonical_record":{"source":{"id":"2504.02193","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-03T00:36:40Z","cross_cats_sorted":[],"title_canon_sha256":"dab600cf8e433591072ae74b0fcde03ad5f2adac61a8b676e2a25209f8f062ff","abstract_canon_sha256":"6f53b93b42543c5b5f424b9ee7081304ff6ffeb7f322a7a3b01ee0e6a11abe94"},"schema_version":"1.0"},"canonical_sha256":"a8e0d6946cf9cffd7878c8f9d9c3e1de4df7c4e2f6b4ea3184ebe7b3b35d03bb","source":{"kind":"arxiv","id":"2504.02193","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.02193","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"arxiv_version","alias_value":"2504.02193v3","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.02193","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"pith_short_12","alias_value":"VDQNNFDM7HH7","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"pith_short_16","alias_value":"VDQNNFDM7HH726DY","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"pith_short_8","alias_value":"VDQNNFDM","created_at":"2026-07-05T11:44:03Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:VDQNNFDM7HH726DYZD45TQ7B3Z","target":"record","payload":{"canonical_record":{"source":{"id":"2504.02193","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-03T00:36:40Z","cross_cats_sorted":[],"title_canon_sha256":"dab600cf8e433591072ae74b0fcde03ad5f2adac61a8b676e2a25209f8f062ff","abstract_canon_sha256":"6f53b93b42543c5b5f424b9ee7081304ff6ffeb7f322a7a3b01ee0e6a11abe94"},"schema_version":"1.0"},"canonical_sha256":"a8e0d6946cf9cffd7878c8f9d9c3e1de4df7c4e2f6b4ea3184ebe7b3b35d03bb","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:44:03.780845Z","signature_b64":"s4OEdm9INGNNf7A6zsPE+pOcDpCjPAa0N2wdm28pDdIb6VHUrovrJlAUOOrYv+sn3SaBKrE6YB4fVK138EeAAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a8e0d6946cf9cffd7878c8f9d9c3e1de4df7c4e2f6b4ea3184ebe7b3b35d03bb","last_reissued_at":"2026-07-05T11:44:03.780297Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:44:03.780297Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2504.02193","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:44:03Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"HYQxyGsRv2zMahgtXs1eCbMXWQk0ZMiUd0AHPrpBIytHvPRh79bq1nHo4ZBkEdUvIBFqgHYcHbUI9yUZ49/8CA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-10T02:55:14.756425Z"},"content_sha256":"6f45010c5475ab407c663576a68ea226fe3c1a6ecd312832875f9d8edaf262a2","schema_version":"1.0","event_id":"sha256:6f45010c5475ab407c663576a68ea226fe3c1a6ecd312832875f9d8edaf262a2"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:VDQNNFDM7HH726DYZD45TQ7B3Z","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"More is Less: The Pitfalls of Multi-Model Synthetic Preference Data in DPO Safety Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Ananth Grama, Bolian Li, David Cho, Junyuan Hong, Runjin Chen, Ruqi Zhang, Tianlong Chen, Yifan Wang, Yihe Deng, Zhangyang Wang","submitted_at":"2025-04-03T00:36:40Z","abstract_excerpt":"Aligning large language models (LLMs) with human values is an increasingly critical step in post-training. Direct Preference Optimization (DPO) has emerged as a simple, yet effective alternative to reinforcement learning from human feedback (RLHF). Synthetic preference data with its low cost and high quality enable effective alignment through single- or multi-model generated preference data. Our study reveals a striking, safety-specific phenomenon associated with DPO alignment: Although multi-model generated data enhances performance on general tasks (ARC, Hellaswag, MMLU, TruthfulQA, Winogran"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.02193","kind":"arxiv","version":3},"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/2504.02193/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":"2026-07-05T11:44:03Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"rKDSx9PhAFTIssnH4/LfHBjPhbu1+AUvitX56LFjjyIxLXFlfYtCXFmVryZWUDUxA8NHADq8MyilD5+uQYnaDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-10T02:55:14.756982Z"},"content_sha256":"ac8a3728ff72f441f4669faa3fe3a916a449607272db2477b276d383618a0d09","schema_version":"1.0","event_id":"sha256:ac8a3728ff72f441f4669faa3fe3a916a449607272db2477b276d383618a0d09"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/VDQNNFDM7HH726DYZD45TQ7B3Z/bundle.json","state_url":"https://pith.science/pith/VDQNNFDM7HH726DYZD45TQ7B3Z/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/VDQNNFDM7HH726DYZD45TQ7B3Z/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":"2026-08-10T02:55:14Z","links":{"resolver":"https://pith.science/pith/VDQNNFDM7HH726DYZD45TQ7B3Z","bundle":"https://pith.science/pith/VDQNNFDM7HH726DYZD45TQ7B3Z/bundle.json","state":"https://pith.science/pith/VDQNNFDM7HH726DYZD45TQ7B3Z/state.json","well_known_bundle":"https://pith.science/.well-known/pith/VDQNNFDM7HH726DYZD45TQ7B3Z/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:VDQNNFDM7HH726DYZD45TQ7B3Z","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":"6f53b93b42543c5b5f424b9ee7081304ff6ffeb7f322a7a3b01ee0e6a11abe94","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-03T00:36:40Z","title_canon_sha256":"dab600cf8e433591072ae74b0fcde03ad5f2adac61a8b676e2a25209f8f062ff"},"schema_version":"1.0","source":{"id":"2504.02193","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.02193","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"arxiv_version","alias_value":"2504.02193v3","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.02193","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"pith_short_12","alias_value":"VDQNNFDM7HH7","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"pith_short_16","alias_value":"VDQNNFDM7HH726DY","created_at":"2026-07-05T11:44:03Z"},{"alias_kind":"pith_short_8","alias_value":"VDQNNFDM","created_at":"2026-07-05T11:44:03Z"}],"graph_snapshots":[{"event_id":"sha256:ac8a3728ff72f441f4669faa3fe3a916a449607272db2477b276d383618a0d09","target":"graph","created_at":"2026-07-05T11:44:03Z","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/2504.02193/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Aligning large language models (LLMs) with human values is an increasingly critical step in post-training. Direct Preference Optimization (DPO) has emerged as a simple, yet effective alternative to reinforcement learning from human feedback (RLHF). Synthetic preference data with its low cost and high quality enable effective alignment through single- or multi-model generated preference data. Our study reveals a striking, safety-specific phenomenon associated with DPO alignment: Although multi-model generated data enhances performance on general tasks (ARC, Hellaswag, MMLU, TruthfulQA, Winogran","authors_text":"Ananth Grama, Bolian Li, David Cho, Junyuan Hong, Runjin Chen, Ruqi Zhang, Tianlong Chen, Yifan Wang, Yihe Deng, Zhangyang Wang","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-03T00:36:40Z","title":"More is Less: The Pitfalls of Multi-Model Synthetic Preference Data in DPO Safety Alignment"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.02193","kind":"arxiv","version":3},"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":"sha256:6f45010c5475ab407c663576a68ea226fe3c1a6ecd312832875f9d8edaf262a2","target":"record","created_at":"2026-07-05T11:44:03Z","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":"6f53b93b42543c5b5f424b9ee7081304ff6ffeb7f322a7a3b01ee0e6a11abe94","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-04-03T00:36:40Z","title_canon_sha256":"dab600cf8e433591072ae74b0fcde03ad5f2adac61a8b676e2a25209f8f062ff"},"schema_version":"1.0","source":{"id":"2504.02193","kind":"arxiv","version":3}},"canonical_sha256":"a8e0d6946cf9cffd7878c8f9d9c3e1de4df7c4e2f6b4ea3184ebe7b3b35d03bb","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a8e0d6946cf9cffd7878c8f9d9c3e1de4df7c4e2f6b4ea3184ebe7b3b35d03bb","first_computed_at":"2026-07-05T11:44:03.780297Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:44:03.780297Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"s4OEdm9INGNNf7A6zsPE+pOcDpCjPAa0N2wdm28pDdIb6VHUrovrJlAUOOrYv+sn3SaBKrE6YB4fVK138EeAAQ==","signature_status":"signed_v1","signed_at":"2026-07-05T11:44:03.780845Z","signed_message":"canonical_sha256_bytes"},"source_id":"2504.02193","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:6f45010c5475ab407c663576a68ea226fe3c1a6ecd312832875f9d8edaf262a2","sha256:ac8a3728ff72f441f4669faa3fe3a916a449607272db2477b276d383618a0d09"],"state_sha256":"79192d79f89fa16b713772390e15216d6848062b8378bd7bcdf4d59e5ced4d19"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Ous+4TiubChqfpkyD6IbJJK6SOplAu0WapFJCIWnCNAmmYzW/BB2oDRDmy7OO0ZVn/NMKsJ6QPNetHbVn39uDQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-10T02:55:14.761765Z","bundle_sha256":"c121c886e4a3377bded463cf8aea8aa7c84e202747425061f5ebe975491e4db4"}}