{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:E3KT6VUYT7NSJBNS6DQRRWNLS6","short_pith_number":"pith:E3KT6VUY","schema_version":"1.0","canonical_sha256":"26d53f56989fdb2485b2f0e118d9ab97b70e7588021140eb7cf27e02d223036b","source":{"kind":"arxiv","id":"2606.17890","version":1},"attestation_state":"computed","paper":{"title":"Dynamic Rollout Editing for Reducing Overthinking in RL-Trained Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Huawei Shen, Jiahao Liu, Jingang Wang, Jingcheng Deng, Liang Pang, Shasha Guo, Shicheng Xu, Wenjie Shi, Xueqi Cheng, Zenghao Duan, Zihao Wei","submitted_at":"2026-06-16T13:10:30Z","abstract_excerpt":"Long-form chain-of-thought reasoning can improve LLM performance on complex tasks, but models often continue generating unnecessary reasoning after a correct answer has emerged. We refer to this behavior as overthinking. We study this phenomenon from the perspective of GRPO-style reinforcement learning (RL) post-training, framing it as a training-time credit-assignment problem rather than merely a decoding-time stopping problem. In rollouts sampled at the onset of GRPO training, we observe that successful trajectories can exhibit a slightly higher degree of overthinking than unsuccessful traje"},"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":"2606.17890","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-06-16T13:10:30Z","cross_cats_sorted":[],"title_canon_sha256":"816220773ba06935060d77d0fe260e9723e5572010a2fcfe752e4f202a6172ac","abstract_canon_sha256":"2687ae0eae80439f4d93d4f3f852847612a5ff9050f5597ee2d15621e706193a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:10:42.782662Z","signature_b64":"vqgybaxC/u1sEfRtyqgsKgM87hU4HCpbLshfs/Rp0Q5Hua2eKkVi3dsHStoN3j+ErJqU7QFRh4VkKSpPEyZwBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"26d53f56989fdb2485b2f0e118d9ab97b70e7588021140eb7cf27e02d223036b","last_reissued_at":"2026-06-19T16:10:42.782235Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:10:42.782235Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dynamic Rollout Editing for Reducing Overthinking in RL-Trained Reasoning Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Huawei Shen, Jiahao Liu, Jingang Wang, Jingcheng Deng, Liang Pang, Shasha Guo, Shicheng Xu, Wenjie Shi, Xueqi Cheng, Zenghao Duan, Zihao Wei","submitted_at":"2026-06-16T13:10:30Z","abstract_excerpt":"Long-form chain-of-thought reasoning can improve LLM performance on complex tasks, but models often continue generating unnecessary reasoning after a correct answer has emerged. We refer to this behavior as overthinking. We study this phenomenon from the perspective of GRPO-style reinforcement learning (RL) post-training, framing it as a training-time credit-assignment problem rather than merely a decoding-time stopping problem. In rollouts sampled at the onset of GRPO training, we observe that successful trajectories can exhibit a slightly higher degree of overthinking than unsuccessful traje"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.17890","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/2606.17890/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":"2606.17890","created_at":"2026-06-19T16:10:42.782302+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.17890v1","created_at":"2026-06-19T16:10:42.782302+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.17890","created_at":"2026-06-19T16:10:42.782302+00:00"},{"alias_kind":"pith_short_12","alias_value":"E3KT6VUYT7NS","created_at":"2026-06-19T16:10:42.782302+00:00"},{"alias_kind":"pith_short_16","alias_value":"E3KT6VUYT7NSJBNS","created_at":"2026-06-19T16:10:42.782302+00:00"},{"alias_kind":"pith_short_8","alias_value":"E3KT6VUY","created_at":"2026-06-19T16:10:42.782302+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/E3KT6VUYT7NSJBNS6DQRRWNLS6","json":"https://pith.science/pith/E3KT6VUYT7NSJBNS6DQRRWNLS6.json","graph_json":"https://pith.science/api/pith-number/E3KT6VUYT7NSJBNS6DQRRWNLS6/graph.json","events_json":"https://pith.science/api/pith-number/E3KT6VUYT7NSJBNS6DQRRWNLS6/events.json","paper":"https://pith.science/paper/E3KT6VUY"},"agent_actions":{"view_html":"https://pith.science/pith/E3KT6VUYT7NSJBNS6DQRRWNLS6","download_json":"https://pith.science/pith/E3KT6VUYT7NSJBNS6DQRRWNLS6.json","view_paper":"https://pith.science/paper/E3KT6VUY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.17890&json=true","fetch_graph":"https://pith.science/api/pith-number/E3KT6VUYT7NSJBNS6DQRRWNLS6/graph.json","fetch_events":"https://pith.science/api/pith-number/E3KT6VUYT7NSJBNS6DQRRWNLS6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/E3KT6VUYT7NSJBNS6DQRRWNLS6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/E3KT6VUYT7NSJBNS6DQRRWNLS6/action/storage_attestation","attest_author":"https://pith.science/pith/E3KT6VUYT7NSJBNS6DQRRWNLS6/action/author_attestation","sign_citation":"https://pith.science/pith/E3KT6VUYT7NSJBNS6DQRRWNLS6/action/citation_signature","submit_replication":"https://pith.science/pith/E3KT6VUYT7NSJBNS6DQRRWNLS6/action/replication_record"}},"created_at":"2026-06-19T16:10:42.782302+00:00","updated_at":"2026-06-19T16:10:42.782302+00:00"}