{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:VC4JEIZLEP36IFLTXU4BL6GVKE","short_pith_number":"pith:VC4JEIZL","schema_version":"1.0","canonical_sha256":"a8b892232b23f7e41573bd3815f8d551260e8156e19991f2444aceefd34a8ec4","source":{"kind":"arxiv","id":"2412.16475","version":1},"attestation_state":"computed","paper":{"title":"When Can Proxies Improve the Sample Complexity of Preference Learning?","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alexander D'Amour, Daniel Augusto de Souza, Matt J. Kusner, Mengyue Yang, Pasquale Minervini, Yuchen Zhu, Zhengyan Shi","submitted_at":"2024-12-21T04:07:17Z","abstract_excerpt":"We address the problem of reward hacking, where maximising a proxy reward does not necessarily increase the true reward. This is a key concern for Large Language Models (LLMs), as they are often fine-tuned on human preferences that may not accurately reflect a true objective. Existing work uses various tricks such as regularisation, tweaks to the reward model, and reward hacking detectors, to limit the influence that such proxy preferences have on a model. Luckily, in many contexts such as medicine, education, and law, a sparse amount of expert data is often available. In these cases, it is 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":"2412.16475","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-12-21T04:07:17Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"33cc06efed653c51ab4ccab24d4034696c2009a204f829222342f299743ad2b7","abstract_canon_sha256":"0420c187b6e04a69b4fa4e2b58de702878352ba7d5b29d212a005fe04c13eac3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:52:48.394381Z","signature_b64":"g1/DN2qMOllH+TLVzKwMRIirNd6hL+gwpPun65UW1jLLaedj1VS45J6qH9tnyI2w06L3kDjPZT9BzouiONrlBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a8b892232b23f7e41573bd3815f8d551260e8156e19991f2444aceefd34a8ec4","last_reissued_at":"2026-07-05T09:52:48.393811Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:52:48.393811Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Can Proxies Improve the Sample Complexity of Preference Learning?","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alexander D'Amour, Daniel Augusto de Souza, Matt J. Kusner, Mengyue Yang, Pasquale Minervini, Yuchen Zhu, Zhengyan Shi","submitted_at":"2024-12-21T04:07:17Z","abstract_excerpt":"We address the problem of reward hacking, where maximising a proxy reward does not necessarily increase the true reward. This is a key concern for Large Language Models (LLMs), as they are often fine-tuned on human preferences that may not accurately reflect a true objective. Existing work uses various tricks such as regularisation, tweaks to the reward model, and reward hacking detectors, to limit the influence that such proxy preferences have on a model. Luckily, in many contexts such as medicine, education, and law, a sparse amount of expert data is often available. In these cases, it is of"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.16475","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/2412.16475/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":"2412.16475","created_at":"2026-07-05T09:52:48.393871+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.16475v1","created_at":"2026-07-05T09:52:48.393871+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.16475","created_at":"2026-07-05T09:52:48.393871+00:00"},{"alias_kind":"pith_short_12","alias_value":"VC4JEIZLEP36","created_at":"2026-07-05T09:52:48.393871+00:00"},{"alias_kind":"pith_short_16","alias_value":"VC4JEIZLEP36IFLT","created_at":"2026-07-05T09:52:48.393871+00:00"},{"alias_kind":"pith_short_8","alias_value":"VC4JEIZL","created_at":"2026-07-05T09:52:48.393871+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/VC4JEIZLEP36IFLTXU4BL6GVKE","json":"https://pith.science/pith/VC4JEIZLEP36IFLTXU4BL6GVKE.json","graph_json":"https://pith.science/api/pith-number/VC4JEIZLEP36IFLTXU4BL6GVKE/graph.json","events_json":"https://pith.science/api/pith-number/VC4JEIZLEP36IFLTXU4BL6GVKE/events.json","paper":"https://pith.science/paper/VC4JEIZL"},"agent_actions":{"view_html":"https://pith.science/pith/VC4JEIZLEP36IFLTXU4BL6GVKE","download_json":"https://pith.science/pith/VC4JEIZLEP36IFLTXU4BL6GVKE.json","view_paper":"https://pith.science/paper/VC4JEIZL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.16475&json=true","fetch_graph":"https://pith.science/api/pith-number/VC4JEIZLEP36IFLTXU4BL6GVKE/graph.json","fetch_events":"https://pith.science/api/pith-number/VC4JEIZLEP36IFLTXU4BL6GVKE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VC4JEIZLEP36IFLTXU4BL6GVKE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VC4JEIZLEP36IFLTXU4BL6GVKE/action/storage_attestation","attest_author":"https://pith.science/pith/VC4JEIZLEP36IFLTXU4BL6GVKE/action/author_attestation","sign_citation":"https://pith.science/pith/VC4JEIZLEP36IFLTXU4BL6GVKE/action/citation_signature","submit_replication":"https://pith.science/pith/VC4JEIZLEP36IFLTXU4BL6GVKE/action/replication_record"}},"created_at":"2026-07-05T09:52:48.393871+00:00","updated_at":"2026-07-05T09:52:48.393871+00:00"}