{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DT7KGEF35RI56XRN43RYREYUPX","short_pith_number":"pith:DT7KGEF3","schema_version":"1.0","canonical_sha256":"1cfea310bbec51df5e2de6e38893147dd14d63f15caef77f1ac67acd177a0317","source":{"kind":"arxiv","id":"2503.21878","version":2},"attestation_state":"computed","paper":{"title":"Is Best-of-N the Best of Them? Coverage, Scaling, and Optimality in Inference-Time Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.AI","authors_text":"Adam Block, Akshay Krishnamurthy, Audrey Huang, Dylan J. Foster, Nan Jiang, Qinghua Liu","submitted_at":"2025-03-27T18:00:08Z","abstract_excerpt":"Inference-time computation offers a powerful axis for scaling the performance of language models. However, naively increasing computation in techniques like Best-of-N sampling can lead to performance degradation due to reward hacking. Toward a theoretical understanding of how to best leverage additional computation, we focus on inference-time alignment, which we formalize as the problem of improving the quality of responses drawn from a pre-trained policy, given a prompt of interest and access to an imperfect reward model. We analyze the performance of inference-time alignment algorithms in te"},"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":"2503.21878","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-03-27T18:00:08Z","cross_cats_sorted":["cs.LG","stat.ML"],"title_canon_sha256":"720afa453ef9992e3b1d2c2b6aa24ac6979557b07ca2e2123ab51b5937bfa66c","abstract_canon_sha256":"edaaca028be5657aa59ee544efb892e4db6aa2b8d6a00f56ae404f76272b1a74"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:45:39.218770Z","signature_b64":"g/4s2e3DmmGIibG7cWnfZAJHhxBPGS1BTghfvYI8OyQYkhRCa4g0+Ud4As38rK+ph1UGRyz5tkUSEDSa3xOLAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1cfea310bbec51df5e2de6e38893147dd14d63f15caef77f1ac67acd177a0317","last_reissued_at":"2026-07-05T10:45:39.218272Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:45:39.218272Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Is Best-of-N the Best of Them? Coverage, Scaling, and Optimality in Inference-Time Alignment","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.AI","authors_text":"Adam Block, Akshay Krishnamurthy, Audrey Huang, Dylan J. Foster, Nan Jiang, Qinghua Liu","submitted_at":"2025-03-27T18:00:08Z","abstract_excerpt":"Inference-time computation offers a powerful axis for scaling the performance of language models. However, naively increasing computation in techniques like Best-of-N sampling can lead to performance degradation due to reward hacking. Toward a theoretical understanding of how to best leverage additional computation, we focus on inference-time alignment, which we formalize as the problem of improving the quality of responses drawn from a pre-trained policy, given a prompt of interest and access to an imperfect reward model. We analyze the performance of inference-time alignment algorithms in te"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.21878","kind":"arxiv","version":2},"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/2503.21878/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":"2503.21878","created_at":"2026-07-05T10:45:39.218340+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.21878v2","created_at":"2026-07-05T10:45:39.218340+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.21878","created_at":"2026-07-05T10:45:39.218340+00:00"},{"alias_kind":"pith_short_12","alias_value":"DT7KGEF35RI5","created_at":"2026-07-05T10:45:39.218340+00:00"},{"alias_kind":"pith_short_16","alias_value":"DT7KGEF35RI56XRN","created_at":"2026-07-05T10:45:39.218340+00:00"},{"alias_kind":"pith_short_8","alias_value":"DT7KGEF3","created_at":"2026-07-05T10:45:39.218340+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.03217","citing_title":"An Asymptotic Theory of Chain-of-Thought in In-Context Learning","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03102","citing_title":"Small RL Controller, Large Language Model: RL-Guided Adaptive Sampling for Test-Time Scaling","ref_index":123,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01667","citing_title":"ATLAS: Agentic Test-time Learning-to-Allocate Scaling","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.31561","citing_title":"What Am I Missing? Question-Answering as Hidden State Probing","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28301","citing_title":"VGB for Masked Diffusion Model: Efficient Test-time Scaling for Reward Satisfaction and Sample Editing","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01075","citing_title":"On the Generalization Gap in Self-Evolving Language Model Reasoning","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2603.27098","citing_title":"Ensemble-Based Uncertainty Estimation for Code Correctness Estimation","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09214","citing_title":"Fast Rates for Offline Contextual Bandits with Forward-KL Regularization under Single-Policy Concentrability","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2604.16453","citing_title":"Sampling for Quality: Training-Free Reward-Guided LLM Decoding via Sequential Monte Carlo","ref_index":7,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DT7KGEF35RI56XRN43RYREYUPX","json":"https://pith.science/pith/DT7KGEF35RI56XRN43RYREYUPX.json","graph_json":"https://pith.science/api/pith-number/DT7KGEF35RI56XRN43RYREYUPX/graph.json","events_json":"https://pith.science/api/pith-number/DT7KGEF35RI56XRN43RYREYUPX/events.json","paper":"https://pith.science/paper/DT7KGEF3"},"agent_actions":{"view_html":"https://pith.science/pith/DT7KGEF35RI56XRN43RYREYUPX","download_json":"https://pith.science/pith/DT7KGEF35RI56XRN43RYREYUPX.json","view_paper":"https://pith.science/paper/DT7KGEF3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.21878&json=true","fetch_graph":"https://pith.science/api/pith-number/DT7KGEF35RI56XRN43RYREYUPX/graph.json","fetch_events":"https://pith.science/api/pith-number/DT7KGEF35RI56XRN43RYREYUPX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DT7KGEF35RI56XRN43RYREYUPX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DT7KGEF35RI56XRN43RYREYUPX/action/storage_attestation","attest_author":"https://pith.science/pith/DT7KGEF35RI56XRN43RYREYUPX/action/author_attestation","sign_citation":"https://pith.science/pith/DT7KGEF35RI56XRN43RYREYUPX/action/citation_signature","submit_replication":"https://pith.science/pith/DT7KGEF35RI56XRN43RYREYUPX/action/replication_record"}},"created_at":"2026-07-05T10:45:39.218340+00:00","updated_at":"2026-07-05T10:45:39.218340+00:00"}