{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7PXHOR7SBD7UOEXXPARCUJ6D3P","short_pith_number":"pith:7PXHOR7S","schema_version":"1.0","canonical_sha256":"fbee7747f208ff4712f778222a27c3dbe0264c5a9459467b17768f0e7636c136","source":{"kind":"arxiv","id":"2501.13011","version":2},"attestation_state":"computed","paper":{"title":"MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Caleb Biddulph, David Elson, David Lindner, Ian Goodfellow, Rohin Shah, Sebastian Farquhar, Vikrant Varma","submitted_at":"2025-01-22T16:53:08Z","abstract_excerpt":"Future advanced AI systems may learn sophisticated strategies through reinforcement learning (RL) that humans cannot understand well enough to safely evaluate. We propose a training method which avoids agents learning undesired multi-step plans that receive high reward (multi-step \"reward hacks\") even if humans are not able to detect that the behaviour is undesired. The method, Myopic Optimization with Non-myopic Approval (MONA), works by combining short-sighted optimization with far-sighted reward. We demonstrate that MONA can prevent multi-step reward hacking that ordinary RL causes, even wi"},"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":"2501.13011","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-01-22T16:53:08Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"7041f472aca480113a09e19aaeb68df5cdaec94cf6302d9f7c7817ea0df0d5d5","abstract_canon_sha256":"538c8b449132aac3b1280bd4b74cea985cda293875ae35a7a3eabad3367d3a1c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:47:20.874811Z","signature_b64":"AKU9W55/goHUrXRTTaGRxChxnqYNXhTnDnCqhjas9raVDOmKwL+MpPlT744xgYK1Utkw62rh7eUjQ+k3rreYDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fbee7747f208ff4712f778222a27c3dbe0264c5a9459467b17768f0e7636c136","last_reissued_at":"2026-07-05T10:47:20.874349Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:47:20.874349Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Caleb Biddulph, David Elson, David Lindner, Ian Goodfellow, Rohin Shah, Sebastian Farquhar, Vikrant Varma","submitted_at":"2025-01-22T16:53:08Z","abstract_excerpt":"Future advanced AI systems may learn sophisticated strategies through reinforcement learning (RL) that humans cannot understand well enough to safely evaluate. We propose a training method which avoids agents learning undesired multi-step plans that receive high reward (multi-step \"reward hacks\") even if humans are not able to detect that the behaviour is undesired. The method, Myopic Optimization with Non-myopic Approval (MONA), works by combining short-sighted optimization with far-sighted reward. We demonstrate that MONA can prevent multi-step reward hacking that ordinary RL causes, even wi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.13011","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/2501.13011/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":"2501.13011","created_at":"2026-07-05T10:47:20.874407+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.13011v2","created_at":"2026-07-05T10:47:20.874407+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.13011","created_at":"2026-07-05T10:47:20.874407+00:00"},{"alias_kind":"pith_short_12","alias_value":"7PXHOR7SBD7U","created_at":"2026-07-05T10:47:20.874407+00:00"},{"alias_kind":"pith_short_16","alias_value":"7PXHOR7SBD7UOEXX","created_at":"2026-07-05T10:47:20.874407+00:00"},{"alias_kind":"pith_short_8","alias_value":"7PXHOR7S","created_at":"2026-07-05T10:47:20.874407+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.30666","citing_title":"Reframing AGI Confrontation with Off Earth Autonomy","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12147","citing_title":"Evaluating Plan Compliance in Autonomous Programming Agents","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13602","citing_title":"Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges","ref_index":181,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7PXHOR7SBD7UOEXXPARCUJ6D3P","json":"https://pith.science/pith/7PXHOR7SBD7UOEXXPARCUJ6D3P.json","graph_json":"https://pith.science/api/pith-number/7PXHOR7SBD7UOEXXPARCUJ6D3P/graph.json","events_json":"https://pith.science/api/pith-number/7PXHOR7SBD7UOEXXPARCUJ6D3P/events.json","paper":"https://pith.science/paper/7PXHOR7S"},"agent_actions":{"view_html":"https://pith.science/pith/7PXHOR7SBD7UOEXXPARCUJ6D3P","download_json":"https://pith.science/pith/7PXHOR7SBD7UOEXXPARCUJ6D3P.json","view_paper":"https://pith.science/paper/7PXHOR7S","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.13011&json=true","fetch_graph":"https://pith.science/api/pith-number/7PXHOR7SBD7UOEXXPARCUJ6D3P/graph.json","fetch_events":"https://pith.science/api/pith-number/7PXHOR7SBD7UOEXXPARCUJ6D3P/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7PXHOR7SBD7UOEXXPARCUJ6D3P/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7PXHOR7SBD7UOEXXPARCUJ6D3P/action/storage_attestation","attest_author":"https://pith.science/pith/7PXHOR7SBD7UOEXXPARCUJ6D3P/action/author_attestation","sign_citation":"https://pith.science/pith/7PXHOR7SBD7UOEXXPARCUJ6D3P/action/citation_signature","submit_replication":"https://pith.science/pith/7PXHOR7SBD7UOEXXPARCUJ6D3P/action/replication_record"}},"created_at":"2026-07-05T10:47:20.874407+00:00","updated_at":"2026-07-05T10:47:20.874407+00:00"}