{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:P5H23MBQ22NRWCMGHCELTWIJ2S","short_pith_number":"pith:P5H23MBQ","schema_version":"1.0","canonical_sha256":"7f4fadb030d69b1b09863888b9d909d4b01c52bcc10ea49ae455fd2731577a1f","source":{"kind":"arxiv","id":"2607.18966","version":1},"attestation_state":"computed","paper":{"title":"Measuring Reward-Seeking via Contrastive Belief Updates","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Alexander Meinke, Axel H{\\o}jmark, Bronson Schoen, Evgenia Nitishinskaya, Felix Hofst\\\"atter, Jason Wolfe, J\\'er\\'emy Scheurer, Theodore Ehrenborg","submitted_at":"2026-07-21T10:57:09Z","abstract_excerpt":"Language models trained with reinforcement learning may learn to optimize the grader's judgment rather than the intended objective. This \"reward-seeking\" is difficult to measure because a model that pursues the grader's judgment and one that pursues the intended objective behave identically whenever the grader rewards the intended behavior. We measure reward-seeking using Contrastive Synthetic Document Finetuning to change a model's beliefs about what the grader rewards, putting those beliefs in conflict with what users or developers want, and measuring the rate at which the model adopts each "},"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":"2607.18966","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-07-21T10:57:09Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"b4ad81fd13a950b828ba651e0f071216f8a9f3bea3f67a3e9cb19e378491b53e","abstract_canon_sha256":"defde27fce502e3e7ebc00d28ba2f59566050630026203fc391d16df36281a81"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-22T01:23:22.216086Z","signature_b64":"b+DHnVD7vmgAJ5zXxL7mZT9P8ABt9ZscMX7ws7pIvhDpIYDKku5tu2wER/LdqKXIwgSWH5zKUYndUK1XXjZYDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7f4fadb030d69b1b09863888b9d909d4b01c52bcc10ea49ae455fd2731577a1f","last_reissued_at":"2026-07-22T01:23:22.215176Z","signature_status":"signed_v1","first_computed_at":"2026-07-22T01:23:22.215176Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Measuring Reward-Seeking via Contrastive Belief Updates","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Alexander Meinke, Axel H{\\o}jmark, Bronson Schoen, Evgenia Nitishinskaya, Felix Hofst\\\"atter, Jason Wolfe, J\\'er\\'emy Scheurer, Theodore Ehrenborg","submitted_at":"2026-07-21T10:57:09Z","abstract_excerpt":"Language models trained with reinforcement learning may learn to optimize the grader's judgment rather than the intended objective. This \"reward-seeking\" is difficult to measure because a model that pursues the grader's judgment and one that pursues the intended objective behave identically whenever the grader rewards the intended behavior. We measure reward-seeking using Contrastive Synthetic Document Finetuning to change a model's beliefs about what the grader rewards, putting those beliefs in conflict with what users or developers want, and measuring the rate at which the model adopts each "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.18966","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/2607.18966/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":"2607.18966","created_at":"2026-07-22T01:23:22.215617+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.18966v1","created_at":"2026-07-22T01:23:22.215617+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.18966","created_at":"2026-07-22T01:23:22.215617+00:00"},{"alias_kind":"pith_short_12","alias_value":"P5H23MBQ22NR","created_at":"2026-07-22T01:23:22.215617+00:00"},{"alias_kind":"pith_short_16","alias_value":"P5H23MBQ22NRWCMG","created_at":"2026-07-22T01:23:22.215617+00:00"},{"alias_kind":"pith_short_8","alias_value":"P5H23MBQ","created_at":"2026-07-22T01:23:22.215617+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/P5H23MBQ22NRWCMGHCELTWIJ2S","json":"https://pith.science/pith/P5H23MBQ22NRWCMGHCELTWIJ2S.json","graph_json":"https://pith.science/api/pith-number/P5H23MBQ22NRWCMGHCELTWIJ2S/graph.json","events_json":"https://pith.science/api/pith-number/P5H23MBQ22NRWCMGHCELTWIJ2S/events.json","paper":"https://pith.science/paper/P5H23MBQ"},"agent_actions":{"view_html":"https://pith.science/pith/P5H23MBQ22NRWCMGHCELTWIJ2S","download_json":"https://pith.science/pith/P5H23MBQ22NRWCMGHCELTWIJ2S.json","view_paper":"https://pith.science/paper/P5H23MBQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.18966&json=true","fetch_graph":"https://pith.science/api/pith-number/P5H23MBQ22NRWCMGHCELTWIJ2S/graph.json","fetch_events":"https://pith.science/api/pith-number/P5H23MBQ22NRWCMGHCELTWIJ2S/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/P5H23MBQ22NRWCMGHCELTWIJ2S/action/timestamp_anchor","attest_storage":"https://pith.science/pith/P5H23MBQ22NRWCMGHCELTWIJ2S/action/storage_attestation","attest_author":"https://pith.science/pith/P5H23MBQ22NRWCMGHCELTWIJ2S/action/author_attestation","sign_citation":"https://pith.science/pith/P5H23MBQ22NRWCMGHCELTWIJ2S/action/citation_signature","submit_replication":"https://pith.science/pith/P5H23MBQ22NRWCMGHCELTWIJ2S/action/replication_record"}},"created_at":"2026-07-22T01:23:22.215617+00:00","updated_at":"2026-07-22T01:23:22.215617+00:00"}