{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:4K5Z7M56SU67IRBVQDGH265YC6","short_pith_number":"pith:4K5Z7M56","schema_version":"1.0","canonical_sha256":"e2bb9fb3be953df4443580cc7d7bb817893a16e54991f4567e9bff059f528c67","source":{"kind":"arxiv","id":"2506.15710","version":1},"attestation_state":"computed","paper":{"title":"RAST: Reasoning Activation in LLMs via Small-model Transfer","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jiawei Han, Minhao Jiang, Siru Ouyang, Xinyu Zhu, Yu Meng, Zilin Xiao","submitted_at":"2025-05-30T17:57:08Z","abstract_excerpt":"Reinforcement learning (RL) has become a powerful approach for improving the reasoning capabilities of large language models (LLMs), as evidenced by recent successes such as OpenAI's o1 and Deepseek-R1. However, applying RL at scale remains intimidatingly resource-intensive, requiring multiple model copies and extensive GPU workloads. On the other hand, while being powerful, recent studies suggest that RL does not fundamentally endow models with new knowledge; rather, it primarily reshapes the model's output distribution to activate reasoning capabilities latent in the base model. Building on "},"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":"2506.15710","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-30T17:57:08Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"4441e90ae07a9f1889f723c47aebcbda9502a677c1597c07f9f74f0137996f49","abstract_canon_sha256":"d57280a51961f0ebbab12d88ac2722ae50815c48b52103178f7bc3dfb39645dc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:23:50.840519Z","signature_b64":"xdMG/lioE9CXanXN3aZkSPb/5Lm5E+CCqo2qn2zvQ48IE+4sNiNVbTkfvQ4ktNCdK7+sNA3unrZuLZ0XaRAeCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e2bb9fb3be953df4443580cc7d7bb817893a16e54991f4567e9bff059f528c67","last_reissued_at":"2026-07-05T11:23:50.840004Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:23:50.840004Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"RAST: Reasoning Activation in LLMs via Small-model Transfer","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Jiawei Han, Minhao Jiang, Siru Ouyang, Xinyu Zhu, Yu Meng, Zilin Xiao","submitted_at":"2025-05-30T17:57:08Z","abstract_excerpt":"Reinforcement learning (RL) has become a powerful approach for improving the reasoning capabilities of large language models (LLMs), as evidenced by recent successes such as OpenAI's o1 and Deepseek-R1. However, applying RL at scale remains intimidatingly resource-intensive, requiring multiple model copies and extensive GPU workloads. On the other hand, while being powerful, recent studies suggest that RL does not fundamentally endow models with new knowledge; rather, it primarily reshapes the model's output distribution to activate reasoning capabilities latent in the base model. Building on "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.15710","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/2506.15710/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":"2506.15710","created_at":"2026-07-05T11:23:50.840065+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.15710v1","created_at":"2026-07-05T11:23:50.840065+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.15710","created_at":"2026-07-05T11:23:50.840065+00:00"},{"alias_kind":"pith_short_12","alias_value":"4K5Z7M56SU67","created_at":"2026-07-05T11:23:50.840065+00:00"},{"alias_kind":"pith_short_16","alias_value":"4K5Z7M56SU67IRBV","created_at":"2026-07-05T11:23:50.840065+00:00"},{"alias_kind":"pith_short_8","alias_value":"4K5Z7M56","created_at":"2026-07-05T11:23:50.840065+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18089","citing_title":"From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning","ref_index":130,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17187","citing_title":"PluRule: A Benchmark for Moderating Pluralistic Communities on Social Media","ref_index":52,"is_internal_anchor":false},{"citing_arxiv_id":"2604.01348","citing_title":"Procedural Knowledge at Scale Improves Reasoning","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2604.06377","citing_title":"The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment","ref_index":47,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4K5Z7M56SU67IRBVQDGH265YC6","json":"https://pith.science/pith/4K5Z7M56SU67IRBVQDGH265YC6.json","graph_json":"https://pith.science/api/pith-number/4K5Z7M56SU67IRBVQDGH265YC6/graph.json","events_json":"https://pith.science/api/pith-number/4K5Z7M56SU67IRBVQDGH265YC6/events.json","paper":"https://pith.science/paper/4K5Z7M56"},"agent_actions":{"view_html":"https://pith.science/pith/4K5Z7M56SU67IRBVQDGH265YC6","download_json":"https://pith.science/pith/4K5Z7M56SU67IRBVQDGH265YC6.json","view_paper":"https://pith.science/paper/4K5Z7M56","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.15710&json=true","fetch_graph":"https://pith.science/api/pith-number/4K5Z7M56SU67IRBVQDGH265YC6/graph.json","fetch_events":"https://pith.science/api/pith-number/4K5Z7M56SU67IRBVQDGH265YC6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4K5Z7M56SU67IRBVQDGH265YC6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4K5Z7M56SU67IRBVQDGH265YC6/action/storage_attestation","attest_author":"https://pith.science/pith/4K5Z7M56SU67IRBVQDGH265YC6/action/author_attestation","sign_citation":"https://pith.science/pith/4K5Z7M56SU67IRBVQDGH265YC6/action/citation_signature","submit_replication":"https://pith.science/pith/4K5Z7M56SU67IRBVQDGH265YC6/action/replication_record"}},"created_at":"2026-07-05T11:23:50.840065+00:00","updated_at":"2026-07-05T11:23:50.840065+00:00"}