{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:EPMR2ASKYCGUJF7XCK5MM4TSEP","short_pith_number":"pith:EPMR2ASK","schema_version":"1.0","canonical_sha256":"23d91d024ac08d4497f712bac6727223e6ed1062e9ccf5cc0b0fcfad822b0bad","source":{"kind":"arxiv","id":"2406.00832","version":3},"attestation_state":"computed","paper":{"title":"BoNBoN Alignment for Large Language Models and the Sweetness of Best-of-n Sampling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Cristina G\\^arbacea, Lin Gui, Victor Veitch","submitted_at":"2024-06-02T18:42:57Z","abstract_excerpt":"This paper concerns the problem of aligning samples from large language models to human preferences using best-of-$n$ sampling, where we draw $n$ samples, rank them, and return the best one. We consider two fundamental problems. First: what is the relationship between best-of-$n$ and approaches to alignment that train LLMs to output samples with a high expected reward (e.g., RLHF or DPO)? To answer this, we embed both the best-of-$n$ distribution and the sampling distributions learned by alignment procedures in a common class of tiltings of the base LLM distribution. We then show that, within "},"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":"2406.00832","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-06-02T18:42:57Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"71723dfd2e5607a9dc05c5ccedae28d1cdd688d6757316149858047738e2231c","abstract_canon_sha256":"b113454aec8f81ec0da5480d7cbd5197e338330ed8da45192261afc09b85e88a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:29:49.215657Z","signature_b64":"dG8itj1gKDYk7hoZnPcw5z+QPHPEN9zlKh82QLbS/eLmnTseEU7KAzUR8ywNyf08DzCUVFvxewO6rUYsJSUHBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"23d91d024ac08d4497f712bac6727223e6ed1062e9ccf5cc0b0fcfad822b0bad","last_reissued_at":"2026-07-05T09:29:49.215186Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:29:49.215186Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"BoNBoN Alignment for Large Language Models and the Sweetness of Best-of-n Sampling","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Cristina G\\^arbacea, Lin Gui, Victor Veitch","submitted_at":"2024-06-02T18:42:57Z","abstract_excerpt":"This paper concerns the problem of aligning samples from large language models to human preferences using best-of-$n$ sampling, where we draw $n$ samples, rank them, and return the best one. We consider two fundamental problems. First: what is the relationship between best-of-$n$ and approaches to alignment that train LLMs to output samples with a high expected reward (e.g., RLHF or DPO)? To answer this, we embed both the best-of-$n$ distribution and the sampling distributions learned by alignment procedures in a common class of tiltings of the base LLM distribution. We then show that, within "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.00832","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/2406.00832/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":"2406.00832","created_at":"2026-07-05T09:29:49.215235+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.00832v3","created_at":"2026-07-05T09:29:49.215235+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.00832","created_at":"2026-07-05T09:29:49.215235+00:00"},{"alias_kind":"pith_short_12","alias_value":"EPMR2ASKYCGU","created_at":"2026-07-05T09:29:49.215235+00:00"},{"alias_kind":"pith_short_16","alias_value":"EPMR2ASKYCGUJF7X","created_at":"2026-07-05T09:29:49.215235+00:00"},{"alias_kind":"pith_short_8","alias_value":"EPMR2ASK","created_at":"2026-07-05T09:29:49.215235+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2505.12741","citing_title":"Language Model Networks: Supervision-Efficient Learning through Dense Communication","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2507.14200","citing_title":"A Scalable Multi-LLM Collaboration System with Retrieval-based Selection and Exploration-Exploitation-Driven Enhancement","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2511.01008","citing_title":"MARS-SQL: A multi-agent reinforcement learning framework for Text-to-SQL","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EPMR2ASKYCGUJF7XCK5MM4TSEP","json":"https://pith.science/pith/EPMR2ASKYCGUJF7XCK5MM4TSEP.json","graph_json":"https://pith.science/api/pith-number/EPMR2ASKYCGUJF7XCK5MM4TSEP/graph.json","events_json":"https://pith.science/api/pith-number/EPMR2ASKYCGUJF7XCK5MM4TSEP/events.json","paper":"https://pith.science/paper/EPMR2ASK"},"agent_actions":{"view_html":"https://pith.science/pith/EPMR2ASKYCGUJF7XCK5MM4TSEP","download_json":"https://pith.science/pith/EPMR2ASKYCGUJF7XCK5MM4TSEP.json","view_paper":"https://pith.science/paper/EPMR2ASK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.00832&json=true","fetch_graph":"https://pith.science/api/pith-number/EPMR2ASKYCGUJF7XCK5MM4TSEP/graph.json","fetch_events":"https://pith.science/api/pith-number/EPMR2ASKYCGUJF7XCK5MM4TSEP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EPMR2ASKYCGUJF7XCK5MM4TSEP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EPMR2ASKYCGUJF7XCK5MM4TSEP/action/storage_attestation","attest_author":"https://pith.science/pith/EPMR2ASKYCGUJF7XCK5MM4TSEP/action/author_attestation","sign_citation":"https://pith.science/pith/EPMR2ASKYCGUJF7XCK5MM4TSEP/action/citation_signature","submit_replication":"https://pith.science/pith/EPMR2ASKYCGUJF7XCK5MM4TSEP/action/replication_record"}},"created_at":"2026-07-05T09:29:49.215235+00:00","updated_at":"2026-07-05T09:29:49.215235+00:00"}