{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:RQ3P7JMD5GGFLKX276QF7BJAZM","short_pith_number":"pith:RQ3P7JMD","schema_version":"1.0","canonical_sha256":"8c36ffa583e98c55aafaffa05f8520cb1cbd289748200b68e1750c143c7abb09","source":{"kind":"arxiv","id":"2605.26315","version":1},"attestation_state":"computed","paper":{"title":"Curriculum Learning for Safety Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chhavi Yadav, Sandeep Kumar, Virginia Smith","submitted_at":"2026-05-25T20:13:06Z","abstract_excerpt":"Direct Preference Optimisation (DPO) is widely used for safety alignment in large language models. However, prior work shows it is brittle and exhibits poor out-of-distribution (OOD) generalisation. In this paper, we investigate whether Curriculum Learning can improve the robustness of DPO-based safety alignment. We propose Staged-Competence, a curriculum-based framework that organises preference data by difficulty, employs competence-based sampling, and progressively updates the reference model during training. Averaged across three model families, Staged-Competence reduces OOD harmful respon"},"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":"2605.26315","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-25T20:13:06Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"1e28a3778344856f0fc975a4a993d20e82100424b2abac7d77422077bb961d32","abstract_canon_sha256":"29f8b6f6a3f669a659c230096f8788db0f5dcdb3e70553b5a260cba5209c1765"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:05:11.605225Z","signature_b64":"iaMEhmwPIUqddLv/LVahy+w9p11qO8r435EdilGtqCL3p+mwLfq27XOPWgkW3ngHQs8vnAr9Y3XTFB8uePLoCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8c36ffa583e98c55aafaffa05f8520cb1cbd289748200b68e1750c143c7abb09","last_reissued_at":"2026-05-27T01:05:11.604395Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:05:11.604395Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Curriculum Learning for Safety Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Chhavi Yadav, Sandeep Kumar, Virginia Smith","submitted_at":"2026-05-25T20:13:06Z","abstract_excerpt":"Direct Preference Optimisation (DPO) is widely used for safety alignment in large language models. However, prior work shows it is brittle and exhibits poor out-of-distribution (OOD) generalisation. In this paper, we investigate whether Curriculum Learning can improve the robustness of DPO-based safety alignment. We propose Staged-Competence, a curriculum-based framework that organises preference data by difficulty, employs competence-based sampling, and progressively updates the reference model during training. Averaged across three model families, Staged-Competence reduces OOD harmful respon"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.26315","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/2605.26315/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":"2605.26315","created_at":"2026-05-27T01:05:11.604531+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.26315v1","created_at":"2026-05-27T01:05:11.604531+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.26315","created_at":"2026-05-27T01:05:11.604531+00:00"},{"alias_kind":"pith_short_12","alias_value":"RQ3P7JMD5GGF","created_at":"2026-05-27T01:05:11.604531+00:00"},{"alias_kind":"pith_short_16","alias_value":"RQ3P7JMD5GGFLKX2","created_at":"2026-05-27T01:05:11.604531+00:00"},{"alias_kind":"pith_short_8","alias_value":"RQ3P7JMD","created_at":"2026-05-27T01:05:11.604531+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/RQ3P7JMD5GGFLKX276QF7BJAZM","json":"https://pith.science/pith/RQ3P7JMD5GGFLKX276QF7BJAZM.json","graph_json":"https://pith.science/api/pith-number/RQ3P7JMD5GGFLKX276QF7BJAZM/graph.json","events_json":"https://pith.science/api/pith-number/RQ3P7JMD5GGFLKX276QF7BJAZM/events.json","paper":"https://pith.science/paper/RQ3P7JMD"},"agent_actions":{"view_html":"https://pith.science/pith/RQ3P7JMD5GGFLKX276QF7BJAZM","download_json":"https://pith.science/pith/RQ3P7JMD5GGFLKX276QF7BJAZM.json","view_paper":"https://pith.science/paper/RQ3P7JMD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.26315&json=true","fetch_graph":"https://pith.science/api/pith-number/RQ3P7JMD5GGFLKX276QF7BJAZM/graph.json","fetch_events":"https://pith.science/api/pith-number/RQ3P7JMD5GGFLKX276QF7BJAZM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RQ3P7JMD5GGFLKX276QF7BJAZM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RQ3P7JMD5GGFLKX276QF7BJAZM/action/storage_attestation","attest_author":"https://pith.science/pith/RQ3P7JMD5GGFLKX276QF7BJAZM/action/author_attestation","sign_citation":"https://pith.science/pith/RQ3P7JMD5GGFLKX276QF7BJAZM/action/citation_signature","submit_replication":"https://pith.science/pith/RQ3P7JMD5GGFLKX276QF7BJAZM/action/replication_record"}},"created_at":"2026-05-27T01:05:11.604531+00:00","updated_at":"2026-05-27T01:05:11.604531+00:00"}