{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:44EID6UY6SXZVW2Q3EAGGCHU7Q","short_pith_number":"pith:44EID6UY","canonical_record":{"source":{"id":"2401.01967","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-01-03T20:26:15Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"807061a19cbaef6dff3337353583b7428fe0f27bd392afdc5278f18a790de53b","abstract_canon_sha256":"56b69a680b9927d068b4c7dc1133f8277db6f8beb1acecbb023112efb1ba1c52"},"schema_version":"1.0"},"canonical_sha256":"e70881fa98f4af9adb50d9006308f4fc24a973ce8169b5430ae4dd51d8c0d6ab","source":{"kind":"arxiv","id":"2401.01967","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2401.01967","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"arxiv_version","alias_value":"2401.01967v1","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.01967","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"pith_short_12","alias_value":"44EID6UY6SXZ","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"pith_short_16","alias_value":"44EID6UY6SXZVW2Q","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"pith_short_8","alias_value":"44EID6UY","created_at":"2026-07-05T07:30:08Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:44EID6UY6SXZVW2Q3EAGGCHU7Q","target":"record","payload":{"canonical_record":{"source":{"id":"2401.01967","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-01-03T20:26:15Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"807061a19cbaef6dff3337353583b7428fe0f27bd392afdc5278f18a790de53b","abstract_canon_sha256":"56b69a680b9927d068b4c7dc1133f8277db6f8beb1acecbb023112efb1ba1c52"},"schema_version":"1.0"},"canonical_sha256":"e70881fa98f4af9adb50d9006308f4fc24a973ce8169b5430ae4dd51d8c0d6ab","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:30:08.881273Z","signature_b64":"7bbvY/j8CMfKaagq+oHjFjMXblHBdLhEUEuG1lvXm0dG1tNIgv9LFBr/DLab59sVF+FiVo8E1iXXdJirqa8bBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e70881fa98f4af9adb50d9006308f4fc24a973ce8169b5430ae4dd51d8c0d6ab","last_reissued_at":"2026-07-05T07:30:08.880579Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:30:08.880579Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2401.01967","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":"2026-07-05T07:30:08Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"11OVWTcqKr7ElU/JDi+jmL5kK/qQAigLbocCu2keyITunbcc18gVM4nD8k71uUn1nBo0H19bUK1WRuW9myXIAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-18T22:48:20.423310Z"},"content_sha256":"34e87e387dc51140ac2c5dfc477bce079f42b5ab9262ed9781360851eeed7227","schema_version":"1.0","event_id":"sha256:34e87e387dc51140ac2c5dfc477bce079f42b5ab9262ed9781360851eeed7227"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:44EID6UY6SXZVW2Q3EAGGCHU7Q","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Andrew Lee, Itamar Pres, Jonathan K. Kummerfeld, Martin Wattenberg, Rada Mihalcea, Xiaoyan Bai","submitted_at":"2024-01-03T20:26:15Z","abstract_excerpt":"While alignment algorithms are now commonly used to tune pre-trained language models towards a user's preferences, we lack explanations for the underlying mechanisms in which models become ``aligned'', thus making it difficult to explain phenomena like jailbreaks. In this work we study a popular algorithm, direct preference optimization (DPO), and the mechanisms by which it reduces toxicity. Namely, we first study how toxicity is represented and elicited in a pre-trained language model, GPT2-medium. We then apply DPO with a carefully crafted pairwise dataset to reduce toxicity. We examine how "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.01967","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/2401.01967/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":"2026-07-05T07:30:08Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"3N15NLdZlDSymbpAQLLt7gXrvYrXdq69VJjoSWXNNZeOnl3YBpei18zlULwyxXm3uB3XqKIy81hA5M1MY5+/Aw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-18T22:48:20.423925Z"},"content_sha256":"965a8ee1af9b204e14ff1c32bec143ce4ec96ddc0771cb0819bf2e0403f6edde","schema_version":"1.0","event_id":"sha256:965a8ee1af9b204e14ff1c32bec143ce4ec96ddc0771cb0819bf2e0403f6edde"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/44EID6UY6SXZVW2Q3EAGGCHU7Q/bundle.json","state_url":"https://pith.science/pith/44EID6UY6SXZVW2Q3EAGGCHU7Q/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/44EID6UY6SXZVW2Q3EAGGCHU7Q/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":"2026-08-18T22:48:20Z","links":{"resolver":"https://pith.science/pith/44EID6UY6SXZVW2Q3EAGGCHU7Q","bundle":"https://pith.science/pith/44EID6UY6SXZVW2Q3EAGGCHU7Q/bundle.json","state":"https://pith.science/pith/44EID6UY6SXZVW2Q3EAGGCHU7Q/state.json","well_known_bundle":"https://pith.science/.well-known/pith/44EID6UY6SXZVW2Q3EAGGCHU7Q/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:44EID6UY6SXZVW2Q3EAGGCHU7Q","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":"56b69a680b9927d068b4c7dc1133f8277db6f8beb1acecbb023112efb1ba1c52","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-01-03T20:26:15Z","title_canon_sha256":"807061a19cbaef6dff3337353583b7428fe0f27bd392afdc5278f18a790de53b"},"schema_version":"1.0","source":{"id":"2401.01967","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2401.01967","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"arxiv_version","alias_value":"2401.01967v1","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.01967","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"pith_short_12","alias_value":"44EID6UY6SXZ","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"pith_short_16","alias_value":"44EID6UY6SXZVW2Q","created_at":"2026-07-05T07:30:08Z"},{"alias_kind":"pith_short_8","alias_value":"44EID6UY","created_at":"2026-07-05T07:30:08Z"}],"graph_snapshots":[{"event_id":"sha256:965a8ee1af9b204e14ff1c32bec143ce4ec96ddc0771cb0819bf2e0403f6edde","target":"graph","created_at":"2026-07-05T07:30:08Z","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/2401.01967/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"While alignment algorithms are now commonly used to tune pre-trained language models towards a user's preferences, we lack explanations for the underlying mechanisms in which models become ``aligned'', thus making it difficult to explain phenomena like jailbreaks. In this work we study a popular algorithm, direct preference optimization (DPO), and the mechanisms by which it reduces toxicity. Namely, we first study how toxicity is represented and elicited in a pre-trained language model, GPT2-medium. We then apply DPO with a carefully crafted pairwise dataset to reduce toxicity. We examine how ","authors_text":"Andrew Lee, Itamar Pres, Jonathan K. Kummerfeld, Martin Wattenberg, Rada Mihalcea, Xiaoyan Bai","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-01-03T20:26:15Z","title":"A Mechanistic Understanding of Alignment Algorithms: A Case Study on DPO and Toxicity"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.01967","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":"sha256:34e87e387dc51140ac2c5dfc477bce079f42b5ab9262ed9781360851eeed7227","target":"record","created_at":"2026-07-05T07:30:08Z","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":"56b69a680b9927d068b4c7dc1133f8277db6f8beb1acecbb023112efb1ba1c52","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-01-03T20:26:15Z","title_canon_sha256":"807061a19cbaef6dff3337353583b7428fe0f27bd392afdc5278f18a790de53b"},"schema_version":"1.0","source":{"id":"2401.01967","kind":"arxiv","version":1}},"canonical_sha256":"e70881fa98f4af9adb50d9006308f4fc24a973ce8169b5430ae4dd51d8c0d6ab","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"e70881fa98f4af9adb50d9006308f4fc24a973ce8169b5430ae4dd51d8c0d6ab","first_computed_at":"2026-07-05T07:30:08.880579Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T07:30:08.880579Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"7bbvY/j8CMfKaagq+oHjFjMXblHBdLhEUEuG1lvXm0dG1tNIgv9LFBr/DLab59sVF+FiVo8E1iXXdJirqa8bBw==","signature_status":"signed_v1","signed_at":"2026-07-05T07:30:08.881273Z","signed_message":"canonical_sha256_bytes"},"source_id":"2401.01967","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:34e87e387dc51140ac2c5dfc477bce079f42b5ab9262ed9781360851eeed7227","sha256:965a8ee1af9b204e14ff1c32bec143ce4ec96ddc0771cb0819bf2e0403f6edde"],"state_sha256":"3cd6290c712be76cac7fc237cff09f4bdd1739e139c33ccca0a445e9de5300e9"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"NJ6CuD1NKZySCxTI9NgV+dAV5nXDwr0PFp0afXXtPY4PxMyvwKFKap8ZXSU/2ms2c/CVBvTVCKw5LGBvrrxZDw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-18T22:48:20.432776Z","bundle_sha256":"cdabb32f4458b4a9d95e82476227f0fe7d8b9b5feab7aee70bf0b6d767cc7c08"}}