{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:YBTKZOIUQT7TVM6NCP4EB2LZE3","short_pith_number":"pith:YBTKZOIU","schema_version":"1.0","canonical_sha256":"c066acb91484ff3ab3cd13f840e97926c2c84d77efbc2fb86df6a206a8ad8266","source":{"kind":"arxiv","id":"2406.05883","version":1},"attestation_state":"computed","paper":{"title":"Information Theoretic Guarantees For Policy Alignment In Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.IT","math.IT","stat.ML"],"primary_cat":"cs.LG","authors_text":"Youssef Mroueh","submitted_at":"2024-06-09T18:41:50Z","abstract_excerpt":"Policy alignment of large language models refers to constrained policy optimization, where the policy is optimized to maximize a reward while staying close to a reference policy with respect to an $f$-divergence such as the $\\mathsf{KL}$ divergence. The best of $n$ alignment policy selects a sample from the reference policy that has the maximum reward among $n$ independent samples. For both cases (policy alignment and best of $n$), recent works showed empirically that the reward improvement of the aligned policy on the reference one scales like $\\sqrt{\\mathsf{KL}}$, with an explicit bound in $"},"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":"2406.05883","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-06-09T18:41:50Z","cross_cats_sorted":["cs.IT","math.IT","stat.ML"],"title_canon_sha256":"f78b627943a0cdb6f2957b1cc214519d51087f5f213826689daed2888c680a8d","abstract_canon_sha256":"3fb903e57e2c0fa4652cd732c7cf229a7045c4e1508cc2c4999bca9b681faee2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:29:34.923227Z","signature_b64":"LpGktnH0TPejM+odIrARjdVA6/TkqHO6NCNfW9fZTZj8aKy4oIk1v6faetUT+AnAmGuSU8fGel3uoIF7YybJDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c066acb91484ff3ab3cd13f840e97926c2c84d77efbc2fb86df6a206a8ad8266","last_reissued_at":"2026-07-05T08:29:34.922807Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:29:34.922807Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Information Theoretic Guarantees For Policy Alignment In Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.IT","math.IT","stat.ML"],"primary_cat":"cs.LG","authors_text":"Youssef Mroueh","submitted_at":"2024-06-09T18:41:50Z","abstract_excerpt":"Policy alignment of large language models refers to constrained policy optimization, where the policy is optimized to maximize a reward while staying close to a reference policy with respect to an $f$-divergence such as the $\\mathsf{KL}$ divergence. The best of $n$ alignment policy selects a sample from the reference policy that has the maximum reward among $n$ independent samples. For both cases (policy alignment and best of $n$), recent works showed empirically that the reward improvement of the aligned policy on the reference one scales like $\\sqrt{\\mathsf{KL}}$, with an explicit bound in $"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.05883","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/2406.05883/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":"2406.05883","created_at":"2026-07-05T08:29:34.922863+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.05883v1","created_at":"2026-07-05T08:29:34.922863+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.05883","created_at":"2026-07-05T08:29:34.922863+00:00"},{"alias_kind":"pith_short_12","alias_value":"YBTKZOIUQT7T","created_at":"2026-07-05T08:29:34.922863+00:00"},{"alias_kind":"pith_short_16","alias_value":"YBTKZOIUQT7TVM6N","created_at":"2026-07-05T08:29:34.922863+00:00"},{"alias_kind":"pith_short_8","alias_value":"YBTKZOIU","created_at":"2026-07-05T08:29:34.922863+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.05913","citing_title":"Best-of-N through the Smoothing Lens: KL Divergence and Regret Analysis","ref_index":38,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YBTKZOIUQT7TVM6NCP4EB2LZE3","json":"https://pith.science/pith/YBTKZOIUQT7TVM6NCP4EB2LZE3.json","graph_json":"https://pith.science/api/pith-number/YBTKZOIUQT7TVM6NCP4EB2LZE3/graph.json","events_json":"https://pith.science/api/pith-number/YBTKZOIUQT7TVM6NCP4EB2LZE3/events.json","paper":"https://pith.science/paper/YBTKZOIU"},"agent_actions":{"view_html":"https://pith.science/pith/YBTKZOIUQT7TVM6NCP4EB2LZE3","download_json":"https://pith.science/pith/YBTKZOIUQT7TVM6NCP4EB2LZE3.json","view_paper":"https://pith.science/paper/YBTKZOIU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.05883&json=true","fetch_graph":"https://pith.science/api/pith-number/YBTKZOIUQT7TVM6NCP4EB2LZE3/graph.json","fetch_events":"https://pith.science/api/pith-number/YBTKZOIUQT7TVM6NCP4EB2LZE3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YBTKZOIUQT7TVM6NCP4EB2LZE3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YBTKZOIUQT7TVM6NCP4EB2LZE3/action/storage_attestation","attest_author":"https://pith.science/pith/YBTKZOIUQT7TVM6NCP4EB2LZE3/action/author_attestation","sign_citation":"https://pith.science/pith/YBTKZOIUQT7TVM6NCP4EB2LZE3/action/citation_signature","submit_replication":"https://pith.science/pith/YBTKZOIUQT7TVM6NCP4EB2LZE3/action/replication_record"}},"created_at":"2026-07-05T08:29:34.922863+00:00","updated_at":"2026-07-05T08:29:34.922863+00:00"}