{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:AKUOQVQN52P2DJ67E2FONQFX7Q","short_pith_number":"pith:AKUOQVQN","schema_version":"1.0","canonical_sha256":"02a8e8560dee9fa1a7df268ae6c0b7fc0c650d2721f7bf85dec6e2d0747d48d0","source":{"kind":"arxiv","id":"2501.19358","version":3},"attestation_state":"computed","paper":{"title":"The Energy Loss Phenomenon in RLHF: A New Perspective on Mitigating Reward Hacking","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dacheng Tao, Lefei Zhang, Liang Ding, Sen Zhang, Yuchun Miao, Yuqi Zhang","submitted_at":"2025-01-31T18:10:53Z","abstract_excerpt":"This work identifies the Energy Loss Phenomenon in Reinforcement Learning from Human Feedback (RLHF) and its connection to reward hacking. Specifically, energy loss in the final layer of a Large Language Model (LLM) gradually increases during the RL process, with an excessive increase in energy loss characterizing reward hacking. Beyond empirical analysis, we further provide a theoretical foundation by proving that, under mild conditions, the increased energy loss reduces the upper bound of contextual relevance in LLMs, which is a critical aspect of reward hacking as the reduced contextual rel"},"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":"2501.19358","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-01-31T18:10:53Z","cross_cats_sorted":[],"title_canon_sha256":"5d44f02811377d4cb6672388eeeb2f70ccf2255126fa1f5b1d4a7aede10fd668","abstract_canon_sha256":"76aff074b5659055ae86044c4066f2650493af99142ff8971d6ed9d78d439499"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:14:03.018776Z","signature_b64":"JMWcqfnbkwmbsv3n6H6BfcdY4qTU+T4J1sZdc/NM+jCs/10H1xO2oz7HPRyMYmzR25P1Y+W78tSHF20o2ZfFCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"02a8e8560dee9fa1a7df268ae6c0b7fc0c650d2721f7bf85dec6e2d0747d48d0","last_reissued_at":"2026-07-05T11:14:03.018222Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:14:03.018222Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Energy Loss Phenomenon in RLHF: A New Perspective on Mitigating Reward Hacking","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dacheng Tao, Lefei Zhang, Liang Ding, Sen Zhang, Yuchun Miao, Yuqi Zhang","submitted_at":"2025-01-31T18:10:53Z","abstract_excerpt":"This work identifies the Energy Loss Phenomenon in Reinforcement Learning from Human Feedback (RLHF) and its connection to reward hacking. Specifically, energy loss in the final layer of a Large Language Model (LLM) gradually increases during the RL process, with an excessive increase in energy loss characterizing reward hacking. Beyond empirical analysis, we further provide a theoretical foundation by proving that, under mild conditions, the increased energy loss reduces the upper bound of contextual relevance in LLMs, which is a critical aspect of reward hacking as the reduced contextual rel"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.19358","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/2501.19358/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":"2501.19358","created_at":"2026-07-05T11:14:03.018295+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.19358v3","created_at":"2026-07-05T11:14:03.018295+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.19358","created_at":"2026-07-05T11:14:03.018295+00:00"},{"alias_kind":"pith_short_12","alias_value":"AKUOQVQN52P2","created_at":"2026-07-05T11:14:03.018295+00:00"},{"alias_kind":"pith_short_16","alias_value":"AKUOQVQN52P2DJ67","created_at":"2026-07-05T11:14:03.018295+00:00"},{"alias_kind":"pith_short_8","alias_value":"AKUOQVQN","created_at":"2026-07-05T11:14:03.018295+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09711","citing_title":"Proxy Reward Internalization and Mechanistic Exploitation: A Learned Precursor to Reward Hacking and Its Generalization","ref_index":255,"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":74,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/AKUOQVQN52P2DJ67E2FONQFX7Q","json":"https://pith.science/pith/AKUOQVQN52P2DJ67E2FONQFX7Q.json","graph_json":"https://pith.science/api/pith-number/AKUOQVQN52P2DJ67E2FONQFX7Q/graph.json","events_json":"https://pith.science/api/pith-number/AKUOQVQN52P2DJ67E2FONQFX7Q/events.json","paper":"https://pith.science/paper/AKUOQVQN"},"agent_actions":{"view_html":"https://pith.science/pith/AKUOQVQN52P2DJ67E2FONQFX7Q","download_json":"https://pith.science/pith/AKUOQVQN52P2DJ67E2FONQFX7Q.json","view_paper":"https://pith.science/paper/AKUOQVQN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.19358&json=true","fetch_graph":"https://pith.science/api/pith-number/AKUOQVQN52P2DJ67E2FONQFX7Q/graph.json","fetch_events":"https://pith.science/api/pith-number/AKUOQVQN52P2DJ67E2FONQFX7Q/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AKUOQVQN52P2DJ67E2FONQFX7Q/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AKUOQVQN52P2DJ67E2FONQFX7Q/action/storage_attestation","attest_author":"https://pith.science/pith/AKUOQVQN52P2DJ67E2FONQFX7Q/action/author_attestation","sign_citation":"https://pith.science/pith/AKUOQVQN52P2DJ67E2FONQFX7Q/action/citation_signature","submit_replication":"https://pith.science/pith/AKUOQVQN52P2DJ67E2FONQFX7Q/action/replication_record"}},"created_at":"2026-07-05T11:14:03.018295+00:00","updated_at":"2026-07-05T11:14:03.018295+00:00"}