{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:UOTXU7ZRE5S6SNLILUFYFP2YSY","short_pith_number":"pith:UOTXU7ZR","schema_version":"1.0","canonical_sha256":"a3a77a7f312765e935685d0b82bf58960641508570adbee92f3efd03c3c7e47e","source":{"kind":"arxiv","id":"2510.21978","version":2},"attestation_state":"computed","paper":{"title":"Beyond Reasoning Gains: Mitigating General-Capability Forgetting in Large Reasoning Models","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Deren Lei, Hoang Phan, Jingyu Zhang, Lijuan Liu, Madian Khabsa, Shengjie Bi, Xianjun Yang, Xiaocheng Tang, Yuanshun Yao","submitted_at":"2025-10-24T19:08:48Z","abstract_excerpt":"Reinforcement learning with verifiable rewards (RLVR) has delivered impressive gains in mathematical and multimodal reasoning and has become a standard post-training paradigm for contemporary language and vision-language models. However, the RLVR recipe introduces a significant risk of capability regression, in which models forget foundational skills after prolonged training without employing regularization strategies. We empirically confirm this concern, observing that open-source reasoning models suffer performance degradation on core capabilities such as perception and faithfulness. While i"},"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":"2510.21978","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/publicdomain/zero/1.0/","primary_cat":"cs.LG","submitted_at":"2025-10-24T19:08:48Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"66925f0b5f2c97c3933cbe564f93637e61192daf908f94f616afe209d5d564d7","abstract_canon_sha256":"b72457433eea86d80aca20623364c9b0a9c5c3f56e92671dd85e54181157c3c2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:12:48.782515Z","signature_b64":"aVKrrUITuxdPqY9oV1sXWRbqG53mjYuNqhhtjMPjt34RcN4goeTtZKrS7lz9zRuoIIjVfS9WLaMq5ahtW3VJCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a3a77a7f312765e935685d0b82bf58960641508570adbee92f3efd03c3c7e47e","last_reissued_at":"2026-06-19T16:12:48.782049Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:12:48.782049Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond Reasoning Gains: Mitigating General-Capability Forgetting in Large Reasoning Models","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Deren Lei, Hoang Phan, Jingyu Zhang, Lijuan Liu, Madian Khabsa, Shengjie Bi, Xianjun Yang, Xiaocheng Tang, Yuanshun Yao","submitted_at":"2025-10-24T19:08:48Z","abstract_excerpt":"Reinforcement learning with verifiable rewards (RLVR) has delivered impressive gains in mathematical and multimodal reasoning and has become a standard post-training paradigm for contemporary language and vision-language models. However, the RLVR recipe introduces a significant risk of capability regression, in which models forget foundational skills after prolonged training without employing regularization strategies. We empirically confirm this concern, observing that open-source reasoning models suffer performance degradation on core capabilities such as perception and faithfulness. While i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.21978","kind":"arxiv","version":2},"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/2510.21978/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":"2510.21978","created_at":"2026-06-19T16:12:48.782105+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.21978v2","created_at":"2026-06-19T16:12:48.782105+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.21978","created_at":"2026-06-19T16:12:48.782105+00:00"},{"alias_kind":"pith_short_12","alias_value":"UOTXU7ZRE5S6","created_at":"2026-06-19T16:12:48.782105+00:00"},{"alias_kind":"pith_short_16","alias_value":"UOTXU7ZRE5S6SNLI","created_at":"2026-06-19T16:12:48.782105+00:00"},{"alias_kind":"pith_short_8","alias_value":"UOTXU7ZR","created_at":"2026-06-19T16:12:48.782105+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2605.08879","citing_title":"Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT","ref_index":23,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12484","citing_title":"Learning, Fast and Slow: Towards LLMs That Adapt Continually","ref_index":43,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08879","citing_title":"Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFT","ref_index":23,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UOTXU7ZRE5S6SNLILUFYFP2YSY","json":"https://pith.science/pith/UOTXU7ZRE5S6SNLILUFYFP2YSY.json","graph_json":"https://pith.science/api/pith-number/UOTXU7ZRE5S6SNLILUFYFP2YSY/graph.json","events_json":"https://pith.science/api/pith-number/UOTXU7ZRE5S6SNLILUFYFP2YSY/events.json","paper":"https://pith.science/paper/UOTXU7ZR"},"agent_actions":{"view_html":"https://pith.science/pith/UOTXU7ZRE5S6SNLILUFYFP2YSY","download_json":"https://pith.science/pith/UOTXU7ZRE5S6SNLILUFYFP2YSY.json","view_paper":"https://pith.science/paper/UOTXU7ZR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.21978&json=true","fetch_graph":"https://pith.science/api/pith-number/UOTXU7ZRE5S6SNLILUFYFP2YSY/graph.json","fetch_events":"https://pith.science/api/pith-number/UOTXU7ZRE5S6SNLILUFYFP2YSY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UOTXU7ZRE5S6SNLILUFYFP2YSY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UOTXU7ZRE5S6SNLILUFYFP2YSY/action/storage_attestation","attest_author":"https://pith.science/pith/UOTXU7ZRE5S6SNLILUFYFP2YSY/action/author_attestation","sign_citation":"https://pith.science/pith/UOTXU7ZRE5S6SNLILUFYFP2YSY/action/citation_signature","submit_replication":"https://pith.science/pith/UOTXU7ZRE5S6SNLILUFYFP2YSY/action/replication_record"}},"created_at":"2026-06-19T16:12:48.782105+00:00","updated_at":"2026-06-19T16:12:48.782105+00:00"}