{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:E5RJHZTBDFJPQTAZ4QVS33XAYB","short_pith_number":"pith:E5RJHZTB","schema_version":"1.0","canonical_sha256":"276293e6611952f84c19e42b2deee0c04a04ae8e145b8f3c0c13e6cd936914c3","source":{"kind":"arxiv","id":"2508.05165","version":1},"attestation_state":"computed","paper":{"title":"Aligning LLMs on a Budget: Inference-Time Alignment with Heuristic Reward Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Hamed Zamani, Kyle H. Wray, Mason Nakamura, Saaduddin Mahmud, Shlomo Zilberstein","submitted_at":"2025-08-07T08:54:27Z","abstract_excerpt":"Aligning LLMs with user preferences is crucial for real-world use but often requires costly fine-tuning or expensive inference, forcing trade-offs between alignment quality and computational cost. Existing inference-time methods typically ignore this balance, focusing solely on the optimized policy's performance. We propose HIA (Heuristic-Guided Inference-time Alignment), a tuning-free, black-box-compatible approach that uses a lightweight prompt optimizer, heuristic reward models, and two-stage filtering to reduce inference calls while preserving alignment quality. On real-world prompt datase"},"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":"2508.05165","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-07T08:54:27Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"110c620fffa4e4d3465474851a33a7f5f30b1be03641de3cc420a7522c3ee771","abstract_canon_sha256":"5e44001f944c4dfe892943566bccb7b39280fe0806d577d27869b5f4c3d50e75"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:50:08.083970Z","signature_b64":"0dblDJiFwk02pa1w/Z+t+wQVOMfK9Tn2CL5VaO44aHlwvBhKaMbfL6S3ntjMU+2lwFsJUNgJjMP/IIMnC9CkCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"276293e6611952f84c19e42b2deee0c04a04ae8e145b8f3c0c13e6cd936914c3","last_reissued_at":"2026-07-05T11:50:08.083405Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:50:08.083405Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Aligning LLMs on a Budget: Inference-Time Alignment with Heuristic Reward Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Hamed Zamani, Kyle H. Wray, Mason Nakamura, Saaduddin Mahmud, Shlomo Zilberstein","submitted_at":"2025-08-07T08:54:27Z","abstract_excerpt":"Aligning LLMs with user preferences is crucial for real-world use but often requires costly fine-tuning or expensive inference, forcing trade-offs between alignment quality and computational cost. Existing inference-time methods typically ignore this balance, focusing solely on the optimized policy's performance. We propose HIA (Heuristic-Guided Inference-time Alignment), a tuning-free, black-box-compatible approach that uses a lightweight prompt optimizer, heuristic reward models, and two-stage filtering to reduce inference calls while preserving alignment quality. On real-world prompt datase"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.05165","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/2508.05165/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":"2508.05165","created_at":"2026-07-05T11:50:08.083530+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.05165v1","created_at":"2026-07-05T11:50:08.083530+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.05165","created_at":"2026-07-05T11:50:08.083530+00:00"},{"alias_kind":"pith_short_12","alias_value":"E5RJHZTBDFJP","created_at":"2026-07-05T11:50:08.083530+00:00"},{"alias_kind":"pith_short_16","alias_value":"E5RJHZTBDFJPQTAZ","created_at":"2026-07-05T11:50:08.083530+00:00"},{"alias_kind":"pith_short_8","alias_value":"E5RJHZTB","created_at":"2026-07-05T11:50:08.083530+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/E5RJHZTBDFJPQTAZ4QVS33XAYB","json":"https://pith.science/pith/E5RJHZTBDFJPQTAZ4QVS33XAYB.json","graph_json":"https://pith.science/api/pith-number/E5RJHZTBDFJPQTAZ4QVS33XAYB/graph.json","events_json":"https://pith.science/api/pith-number/E5RJHZTBDFJPQTAZ4QVS33XAYB/events.json","paper":"https://pith.science/paper/E5RJHZTB"},"agent_actions":{"view_html":"https://pith.science/pith/E5RJHZTBDFJPQTAZ4QVS33XAYB","download_json":"https://pith.science/pith/E5RJHZTBDFJPQTAZ4QVS33XAYB.json","view_paper":"https://pith.science/paper/E5RJHZTB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.05165&json=true","fetch_graph":"https://pith.science/api/pith-number/E5RJHZTBDFJPQTAZ4QVS33XAYB/graph.json","fetch_events":"https://pith.science/api/pith-number/E5RJHZTBDFJPQTAZ4QVS33XAYB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/E5RJHZTBDFJPQTAZ4QVS33XAYB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/E5RJHZTBDFJPQTAZ4QVS33XAYB/action/storage_attestation","attest_author":"https://pith.science/pith/E5RJHZTBDFJPQTAZ4QVS33XAYB/action/author_attestation","sign_citation":"https://pith.science/pith/E5RJHZTBDFJPQTAZ4QVS33XAYB/action/citation_signature","submit_replication":"https://pith.science/pith/E5RJHZTBDFJPQTAZ4QVS33XAYB/action/replication_record"}},"created_at":"2026-07-05T11:50:08.083530+00:00","updated_at":"2026-07-05T11:50:08.083530+00:00"}