{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:X4275CXQSYDHADXRXDP6VY4L77","short_pith_number":"pith:X4275CXQ","schema_version":"1.0","canonical_sha256":"bf35fe8af09606700ef1b8dfeae38bffee4f6bbabdc315aa4aaacebc440ed4c9","source":{"kind":"arxiv","id":"2404.01730","version":1},"attestation_state":"computed","paper":{"title":"Asymptotics of Language Model Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.IT","math.IT","stat.ML"],"primary_cat":"cs.LG","authors_text":"Ahmad Beirami, Ananda Theertha Suresh, Joy Qiping Yang, Salman Salamatian, Ziteng Sun","submitted_at":"2024-04-02T08:40:07Z","abstract_excerpt":"Let $p$ denote a generative language model. Let $r$ denote a reward model that returns a scalar that captures the degree at which a draw from $p$ is preferred. The goal of language model alignment is to alter $p$ to a new distribution $\\phi$ that results in a higher expected reward while keeping $\\phi$ close to $p.$ A popular alignment method is the KL-constrained reinforcement learning (RL), which chooses a distribution $\\phi_\\Delta$ that maximizes $E_{\\phi_{\\Delta}} r(y)$ subject to a relative entropy constraint $KL(\\phi_\\Delta || p) \\leq \\Delta.$ Another simple alignment method is best-of-$"},"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":"2404.01730","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-04-02T08:40:07Z","cross_cats_sorted":["cs.IT","math.IT","stat.ML"],"title_canon_sha256":"8dc3f2d90b66ffdf710896d306b59e3687f09670d1358953aa40b9ffa61cf176","abstract_canon_sha256":"6b06e441d66c1e1b848b1eb51290d264627b90c3a92d2896d7a819ac940670c7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:03:18.891637Z","signature_b64":"XpZo3bdsE0EV1bZvHkLlVGwNKPeYOfLDq/hqYM1gu+QR64QMwn23rpYTPERllIVeQSl+RUTsX378eobHn84gDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bf35fe8af09606700ef1b8dfeae38bffee4f6bbabdc315aa4aaacebc440ed4c9","last_reissued_at":"2026-07-05T08:03:18.891229Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:03:18.891229Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Asymptotics of Language Model Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.IT","math.IT","stat.ML"],"primary_cat":"cs.LG","authors_text":"Ahmad Beirami, Ananda Theertha Suresh, Joy Qiping Yang, Salman Salamatian, Ziteng Sun","submitted_at":"2024-04-02T08:40:07Z","abstract_excerpt":"Let $p$ denote a generative language model. Let $r$ denote a reward model that returns a scalar that captures the degree at which a draw from $p$ is preferred. The goal of language model alignment is to alter $p$ to a new distribution $\\phi$ that results in a higher expected reward while keeping $\\phi$ close to $p.$ A popular alignment method is the KL-constrained reinforcement learning (RL), which chooses a distribution $\\phi_\\Delta$ that maximizes $E_{\\phi_{\\Delta}} r(y)$ subject to a relative entropy constraint $KL(\\phi_\\Delta || p) \\leq \\Delta.$ Another simple alignment method is best-of-$"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.01730","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/2404.01730/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":"2404.01730","created_at":"2026-07-05T08:03:18.891280+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.01730v1","created_at":"2026-07-05T08:03:18.891280+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.01730","created_at":"2026-07-05T08:03:18.891280+00:00"},{"alias_kind":"pith_short_12","alias_value":"X4275CXQSYDH","created_at":"2026-07-05T08:03:18.891280+00:00"},{"alias_kind":"pith_short_16","alias_value":"X4275CXQSYDHADXR","created_at":"2026-07-05T08:03:18.891280+00:00"},{"alias_kind":"pith_short_8","alias_value":"X4275CXQ","created_at":"2026-07-05T08:03:18.891280+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2412.19792","citing_title":"InfAlign: Inference-aware language model alignment","ref_index":64,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/X4275CXQSYDHADXRXDP6VY4L77","json":"https://pith.science/pith/X4275CXQSYDHADXRXDP6VY4L77.json","graph_json":"https://pith.science/api/pith-number/X4275CXQSYDHADXRXDP6VY4L77/graph.json","events_json":"https://pith.science/api/pith-number/X4275CXQSYDHADXRXDP6VY4L77/events.json","paper":"https://pith.science/paper/X4275CXQ"},"agent_actions":{"view_html":"https://pith.science/pith/X4275CXQSYDHADXRXDP6VY4L77","download_json":"https://pith.science/pith/X4275CXQSYDHADXRXDP6VY4L77.json","view_paper":"https://pith.science/paper/X4275CXQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.01730&json=true","fetch_graph":"https://pith.science/api/pith-number/X4275CXQSYDHADXRXDP6VY4L77/graph.json","fetch_events":"https://pith.science/api/pith-number/X4275CXQSYDHADXRXDP6VY4L77/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/X4275CXQSYDHADXRXDP6VY4L77/action/timestamp_anchor","attest_storage":"https://pith.science/pith/X4275CXQSYDHADXRXDP6VY4L77/action/storage_attestation","attest_author":"https://pith.science/pith/X4275CXQSYDHADXRXDP6VY4L77/action/author_attestation","sign_citation":"https://pith.science/pith/X4275CXQSYDHADXRXDP6VY4L77/action/citation_signature","submit_replication":"https://pith.science/pith/X4275CXQSYDHADXRXDP6VY4L77/action/replication_record"}},"created_at":"2026-07-05T08:03:18.891280+00:00","updated_at":"2026-07-05T08:03:18.891280+00:00"}