{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:NRI6OEDOGBDPYWFSMQRFGGKTRX","short_pith_number":"pith:NRI6OEDO","schema_version":"1.0","canonical_sha256":"6c51e7106e3046fc58b264225319538ddd8cc9e9f29b2deea091b3493503a5b4","source":{"kind":"arxiv","id":"2402.06627","version":3},"attestation_state":"computed","paper":{"title":"Feedback Loops With Language Models Drive In-Context Reward Hacking","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alexander Pan, Erik Jones, Jacob Steinhardt, Meena Jagadeesan","submitted_at":"2024-02-09T18:59:29Z","abstract_excerpt":"Language models influence the external world: they query APIs that read and write to web pages, generate content that shapes human behavior, and run system commands as autonomous agents. These interactions form feedback loops: LLM outputs affect the world, which in turn affect subsequent LLM outputs. In this work, we show that feedback loops can cause in-context reward hacking (ICRH), where the LLM at test-time optimizes a (potentially implicit) objective but creates negative side effects in the process. For example, consider an LLM agent deployed to increase Twitter engagement; the LLM may re"},"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":"2402.06627","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-09T18:59:29Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"7b37f06c0552076f15c256e032c946bf294cc507e6460df0c13c0f351b999d6d","abstract_canon_sha256":"7d3f5aacb68a93a40c5d5d9249fc970fade483a6398c897fb34cb60121aeb983"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:28:37.979172Z","signature_b64":"BOY6m0CBfh/euhejS+nhhaFoBd1bXoUWktq9/d9M4WROLW1rWL7RNE9oYrpCr3OzY0PT+LlNyPe2/xMPCVd4CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6c51e7106e3046fc58b264225319538ddd8cc9e9f29b2deea091b3493503a5b4","last_reissued_at":"2026-07-05T08:28:37.978707Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:28:37.978707Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Feedback Loops With Language Models Drive In-Context Reward Hacking","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alexander Pan, Erik Jones, Jacob Steinhardt, Meena Jagadeesan","submitted_at":"2024-02-09T18:59:29Z","abstract_excerpt":"Language models influence the external world: they query APIs that read and write to web pages, generate content that shapes human behavior, and run system commands as autonomous agents. These interactions form feedback loops: LLM outputs affect the world, which in turn affect subsequent LLM outputs. In this work, we show that feedback loops can cause in-context reward hacking (ICRH), where the LLM at test-time optimizes a (potentially implicit) objective but creates negative side effects in the process. For example, consider an LLM agent deployed to increase Twitter engagement; the LLM may re"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.06627","kind":"arxiv","version":3},"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/2402.06627/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":"2402.06627","created_at":"2026-07-05T08:28:37.978767+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.06627v3","created_at":"2026-07-05T08:28:37.978767+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.06627","created_at":"2026-07-05T08:28:37.978767+00:00"},{"alias_kind":"pith_short_12","alias_value":"NRI6OEDOGBDP","created_at":"2026-07-05T08:28:37.978767+00:00"},{"alias_kind":"pith_short_16","alias_value":"NRI6OEDOGBDPYWFS","created_at":"2026-07-05T08:28:37.978767+00:00"},{"alias_kind":"pith_short_8","alias_value":"NRI6OEDO","created_at":"2026-07-05T08:28:37.978767+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2404.13076","citing_title":"LLM Evaluators Recognize and Favor Their Own Generations","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2506.12382","citing_title":"Exploring the Secondary Risks of Large Language Models","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14220","citing_title":"Diagnosing Training Inference Mismatch in LLM Reinforcement Learning","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12673","citing_title":"Do Androids Dream of Breaking the Game? Systematically Auditing AI Agent Benchmarks with BenchJack","ref_index":41,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02585","citing_title":"Mitigating LLM biases toward spurious social contexts using direct preference optimization","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2504.07615","citing_title":"VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model","ref_index":40,"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":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NRI6OEDOGBDPYWFSMQRFGGKTRX","json":"https://pith.science/pith/NRI6OEDOGBDPYWFSMQRFGGKTRX.json","graph_json":"https://pith.science/api/pith-number/NRI6OEDOGBDPYWFSMQRFGGKTRX/graph.json","events_json":"https://pith.science/api/pith-number/NRI6OEDOGBDPYWFSMQRFGGKTRX/events.json","paper":"https://pith.science/paper/NRI6OEDO"},"agent_actions":{"view_html":"https://pith.science/pith/NRI6OEDOGBDPYWFSMQRFGGKTRX","download_json":"https://pith.science/pith/NRI6OEDOGBDPYWFSMQRFGGKTRX.json","view_paper":"https://pith.science/paper/NRI6OEDO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.06627&json=true","fetch_graph":"https://pith.science/api/pith-number/NRI6OEDOGBDPYWFSMQRFGGKTRX/graph.json","fetch_events":"https://pith.science/api/pith-number/NRI6OEDOGBDPYWFSMQRFGGKTRX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NRI6OEDOGBDPYWFSMQRFGGKTRX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NRI6OEDOGBDPYWFSMQRFGGKTRX/action/storage_attestation","attest_author":"https://pith.science/pith/NRI6OEDOGBDPYWFSMQRFGGKTRX/action/author_attestation","sign_citation":"https://pith.science/pith/NRI6OEDOGBDPYWFSMQRFGGKTRX/action/citation_signature","submit_replication":"https://pith.science/pith/NRI6OEDOGBDPYWFSMQRFGGKTRX/action/replication_record"}},"created_at":"2026-07-05T08:28:37.978767+00:00","updated_at":"2026-07-05T08:28:37.978767+00:00"}