{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZNTNUA2ZYABWDWWYIBEZWK3W3M","short_pith_number":"pith:ZNTNUA2Z","schema_version":"1.0","canonical_sha256":"cb66da0359c00361dad840499b2b76db06b9e7f10dd4a88de813c5b55cb04f8f","source":{"kind":"arxiv","id":"2402.08005","version":1},"attestation_state":"computed","paper":{"title":"Refined Direct Preference Optimization with Synthetic Data for Behavioral Alignment of LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"V\\'ictor Gallego","submitted_at":"2024-02-12T19:10:13Z","abstract_excerpt":"In this paper, we introduce \\emph{refined Direct Preference Optimization} (rDPO), a method for improving the behavioral alignment of Large Language Models (LLMs) without the need for human-annotated data. The method involves creating synthetic data using self-critique prompting by a teacher LLM and then utilising a generalized DPO loss function to distil to a student LLM. The loss function incorporates an additional external reward model to improve the quality of synthetic data, making rDPO robust to potential noise in the synthetic dataset. rDPO is shown to be effective in a diverse set of be"},"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":"2402.08005","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-02-12T19:10:13Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"322a6139774c34e13d5e7ec6f01a8b95adff823a9bf91db974d574124e85bf72","abstract_canon_sha256":"d2958e7d20a3a1af4094bee8e004de335d737d370eea92883093bdf3ca9dbec3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:44:25.634958Z","signature_b64":"pTdCjlYUzqByeAbFZupE8iSX1mqjPS0cKfGxxK1z6XzhFXW8FEZNcYNkq+2RynsH8rZnKbwzM+Ru0S5XfdqFDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cb66da0359c00361dad840499b2b76db06b9e7f10dd4a88de813c5b55cb04f8f","last_reissued_at":"2026-07-05T07:44:25.634466Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:44:25.634466Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Refined Direct Preference Optimization with Synthetic Data for Behavioral Alignment of LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"V\\'ictor Gallego","submitted_at":"2024-02-12T19:10:13Z","abstract_excerpt":"In this paper, we introduce \\emph{refined Direct Preference Optimization} (rDPO), a method for improving the behavioral alignment of Large Language Models (LLMs) without the need for human-annotated data. The method involves creating synthetic data using self-critique prompting by a teacher LLM and then utilising a generalized DPO loss function to distil to a student LLM. The loss function incorporates an additional external reward model to improve the quality of synthetic data, making rDPO robust to potential noise in the synthetic dataset. rDPO is shown to be effective in a diverse set of be"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.08005","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/2402.08005/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":"2402.08005","created_at":"2026-07-05T07:44:25.634526+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.08005v1","created_at":"2026-07-05T07:44:25.634526+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.08005","created_at":"2026-07-05T07:44:25.634526+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZNTNUA2ZYABW","created_at":"2026-07-05T07:44:25.634526+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZNTNUA2ZYABWDWWY","created_at":"2026-07-05T07:44:25.634526+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZNTNUA2Z","created_at":"2026-07-05T07:44:25.634526+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23244","citing_title":"Convex Optimization for Alignment and Preference Learning on a Single GPU","ref_index":81,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZNTNUA2ZYABWDWWYIBEZWK3W3M","json":"https://pith.science/pith/ZNTNUA2ZYABWDWWYIBEZWK3W3M.json","graph_json":"https://pith.science/api/pith-number/ZNTNUA2ZYABWDWWYIBEZWK3W3M/graph.json","events_json":"https://pith.science/api/pith-number/ZNTNUA2ZYABWDWWYIBEZWK3W3M/events.json","paper":"https://pith.science/paper/ZNTNUA2Z"},"agent_actions":{"view_html":"https://pith.science/pith/ZNTNUA2ZYABWDWWYIBEZWK3W3M","download_json":"https://pith.science/pith/ZNTNUA2ZYABWDWWYIBEZWK3W3M.json","view_paper":"https://pith.science/paper/ZNTNUA2Z","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.08005&json=true","fetch_graph":"https://pith.science/api/pith-number/ZNTNUA2ZYABWDWWYIBEZWK3W3M/graph.json","fetch_events":"https://pith.science/api/pith-number/ZNTNUA2ZYABWDWWYIBEZWK3W3M/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZNTNUA2ZYABWDWWYIBEZWK3W3M/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZNTNUA2ZYABWDWWYIBEZWK3W3M/action/storage_attestation","attest_author":"https://pith.science/pith/ZNTNUA2ZYABWDWWYIBEZWK3W3M/action/author_attestation","sign_citation":"https://pith.science/pith/ZNTNUA2ZYABWDWWYIBEZWK3W3M/action/citation_signature","submit_replication":"https://pith.science/pith/ZNTNUA2ZYABWDWWYIBEZWK3W3M/action/replication_record"}},"created_at":"2026-07-05T07:44:25.634526+00:00","updated_at":"2026-07-05T07:44:25.634526+00:00"}