{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZHCSITUN6G6BSW2FNILRH2DGLY","short_pith_number":"pith:ZHCSITUN","schema_version":"1.0","canonical_sha256":"c9c5244e8df1bc195b456a1713e8665e249e18638f25540a169e8a29bbdef92f","source":{"kind":"arxiv","id":"2608.01743","version":1},"attestation_state":"computed","paper":{"title":"Toward Plasticity-Preserving KL Regularization for Capability Retention in LLM Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Guojun Yin, Jiajun Chai, Li Wang, Tianhao Peng, Wei Lin, Xiaodong Lu, Xiaohan Wang","submitted_at":"2026-08-03T06:10:33Z","abstract_excerpt":"Reinforcement learning (RL) has become a central paradigm for large language model (LLM) post-training, but optimization toward new objectives can degrade capabilities already present in the base model. KL regularization is widely used to mitigate such forgetting by constraining policy drift toward a reference model. However, standard full-policy KL regularization constrains the entire response distribution and may unnecessarily restrict exploration and target-task learning. This raises a natural question: can a more precise constraint preserve existing capabilities while minimizing interferen"},"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":"2608.01743","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-08-03T06:10:33Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"73173a3ddf6792d7efaed597af83fb46ff34c72214daf4be345fd8a2075e8e2c","abstract_canon_sha256":"af9741c2f25483172e8a9b2850f2c46c792756ff8060373637a544d3cfce475d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-04T02:06:40.620613Z","signature_b64":"WAoKHjj6dcbtP1zq5YAIR4OsJOr/KnjDnJqfW547cMzoNT7kx1AIcfHYHkTPlX9RlNHd5yD8q4r+Kslye7D+CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c9c5244e8df1bc195b456a1713e8665e249e18638f25540a169e8a29bbdef92f","last_reissued_at":"2026-08-04T02:06:40.619008Z","signature_status":"signed_v1","first_computed_at":"2026-08-04T02:06:40.619008Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Toward Plasticity-Preserving KL Regularization for Capability Retention in LLM Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Guojun Yin, Jiajun Chai, Li Wang, Tianhao Peng, Wei Lin, Xiaodong Lu, Xiaohan Wang","submitted_at":"2026-08-03T06:10:33Z","abstract_excerpt":"Reinforcement learning (RL) has become a central paradigm for large language model (LLM) post-training, but optimization toward new objectives can degrade capabilities already present in the base model. KL regularization is widely used to mitigate such forgetting by constraining policy drift toward a reference model. However, standard full-policy KL regularization constrains the entire response distribution and may unnecessarily restrict exploration and target-task learning. This raises a natural question: can a more precise constraint preserve existing capabilities while minimizing interferen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.01743","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/2608.01743/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":"2608.01743","created_at":"2026-08-04T02:06:40.620528+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.01743v1","created_at":"2026-08-04T02:06:40.620528+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.01743","created_at":"2026-08-04T02:06:40.620528+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZHCSITUN6G6B","created_at":"2026-08-04T02:06:40.620528+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZHCSITUN6G6BSW2F","created_at":"2026-08-04T02:06:40.620528+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZHCSITUN","created_at":"2026-08-04T02:06:40.620528+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/ZHCSITUN6G6BSW2FNILRH2DGLY","json":"https://pith.science/pith/ZHCSITUN6G6BSW2FNILRH2DGLY.json","graph_json":"https://pith.science/api/pith-number/ZHCSITUN6G6BSW2FNILRH2DGLY/graph.json","events_json":"https://pith.science/api/pith-number/ZHCSITUN6G6BSW2FNILRH2DGLY/events.json","paper":"https://pith.science/paper/ZHCSITUN"},"agent_actions":{"view_html":"https://pith.science/pith/ZHCSITUN6G6BSW2FNILRH2DGLY","download_json":"https://pith.science/pith/ZHCSITUN6G6BSW2FNILRH2DGLY.json","view_paper":"https://pith.science/paper/ZHCSITUN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.01743&json=true","fetch_graph":"https://pith.science/api/pith-number/ZHCSITUN6G6BSW2FNILRH2DGLY/graph.json","fetch_events":"https://pith.science/api/pith-number/ZHCSITUN6G6BSW2FNILRH2DGLY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZHCSITUN6G6BSW2FNILRH2DGLY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZHCSITUN6G6BSW2FNILRH2DGLY/action/storage_attestation","attest_author":"https://pith.science/pith/ZHCSITUN6G6BSW2FNILRH2DGLY/action/author_attestation","sign_citation":"https://pith.science/pith/ZHCSITUN6G6BSW2FNILRH2DGLY/action/citation_signature","submit_replication":"https://pith.science/pith/ZHCSITUN6G6BSW2FNILRH2DGLY/action/replication_record"}},"created_at":"2026-08-04T02:06:40.620528+00:00","updated_at":"2026-08-04T02:06:40.620528+00:00"}