{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ABGZODK2EMPRB65444474VRHR4","short_pith_number":"pith:ABGZODK2","schema_version":"1.0","canonical_sha256":"004d970d5a231f10fbbce739fe56278f1ebcab378da6966ebb4504964ee26068","source":{"kind":"arxiv","id":"2502.20122","version":3},"attestation_state":"computed","paper":{"title":"Self-Training Elicits Concise Reasoning in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Namgyu Ho, Seo Hyun Kim, Se-Young Yun, Tergel Munkhbat, Yongjin Yang, Yujin Kim","submitted_at":"2025-02-27T14:14:50Z","abstract_excerpt":"Chain-of-thought (CoT) reasoning has enabled large language models (LLMs) to utilize additional computation through intermediate tokens to solve complex tasks. However, we posit that typical reasoning traces contain many redundant tokens, incurring extraneous inference costs. Upon examination of the output distribution of current LLMs, we find evidence on their latent ability to reason more concisely, relative to their default behavior. To elicit this capability, we propose simple fine-tuning methods which leverage self-generated concise reasoning paths obtained by best-of-N sampling and few-s"},"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":"2502.20122","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-27T14:14:50Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"6c1c17d812aad95bb3a5c024e7441ee2f53c6ddf95b54e684f871279131362fc","abstract_canon_sha256":"42ea70d1469dd6e51611424b484f1ca324ef3698b627170e8f39010b79830708"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:03.867970Z","signature_b64":"3OcI4cCsnRcaxOWteX6csSlUR85ZKJOB+EC/3ky6EoIPqWk36A6wyodkJSrZaxbzGIbM9OEtJepx2ki6wwrkBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"004d970d5a231f10fbbce739fe56278f1ebcab378da6966ebb4504964ee26068","last_reissued_at":"2026-07-05T11:19:03.867451Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:03.867451Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Self-Training Elicits Concise Reasoning in Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Namgyu Ho, Seo Hyun Kim, Se-Young Yun, Tergel Munkhbat, Yongjin Yang, Yujin Kim","submitted_at":"2025-02-27T14:14:50Z","abstract_excerpt":"Chain-of-thought (CoT) reasoning has enabled large language models (LLMs) to utilize additional computation through intermediate tokens to solve complex tasks. However, we posit that typical reasoning traces contain many redundant tokens, incurring extraneous inference costs. Upon examination of the output distribution of current LLMs, we find evidence on their latent ability to reason more concisely, relative to their default behavior. To elicit this capability, we propose simple fine-tuning methods which leverage self-generated concise reasoning paths obtained by best-of-N sampling and few-s"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.20122","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/2502.20122/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":"2502.20122","created_at":"2026-07-05T11:19:03.867512+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.20122v3","created_at":"2026-07-05T11:19:03.867512+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.20122","created_at":"2026-07-05T11:19:03.867512+00:00"},{"alias_kind":"pith_short_12","alias_value":"ABGZODK2EMPR","created_at":"2026-07-05T11:19:03.867512+00:00"},{"alias_kind":"pith_short_16","alias_value":"ABGZODK2EMPRB654","created_at":"2026-07-05T11:19:03.867512+00:00"},{"alias_kind":"pith_short_8","alias_value":"ABGZODK2","created_at":"2026-07-05T11:19:03.867512+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.07108","citing_title":"DyCon: Dynamic Reasoning Control via Evolving Difficulty Modeling","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01168","citing_title":"Thinking Economically: A Hierarchical Framework for Adaptive-Complexity Reasoning in LLMs","ref_index":48,"is_internal_anchor":false},{"citing_arxiv_id":"2503.16419","citing_title":"Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models","ref_index":133,"is_internal_anchor":false},{"citing_arxiv_id":"2502.17419","citing_title":"From System 1 to System 2: A Survey of Reasoning Large Language Models","ref_index":202,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06165","citing_title":"Post Reasoning: Improving the Performance of Non-Thinking Models at No Cost","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01111","citing_title":"When Less is Enough: Efficient Inference via Collaborative Reasoning","ref_index":29,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ABGZODK2EMPRB65444474VRHR4","json":"https://pith.science/pith/ABGZODK2EMPRB65444474VRHR4.json","graph_json":"https://pith.science/api/pith-number/ABGZODK2EMPRB65444474VRHR4/graph.json","events_json":"https://pith.science/api/pith-number/ABGZODK2EMPRB65444474VRHR4/events.json","paper":"https://pith.science/paper/ABGZODK2"},"agent_actions":{"view_html":"https://pith.science/pith/ABGZODK2EMPRB65444474VRHR4","download_json":"https://pith.science/pith/ABGZODK2EMPRB65444474VRHR4.json","view_paper":"https://pith.science/paper/ABGZODK2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.20122&json=true","fetch_graph":"https://pith.science/api/pith-number/ABGZODK2EMPRB65444474VRHR4/graph.json","fetch_events":"https://pith.science/api/pith-number/ABGZODK2EMPRB65444474VRHR4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ABGZODK2EMPRB65444474VRHR4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ABGZODK2EMPRB65444474VRHR4/action/storage_attestation","attest_author":"https://pith.science/pith/ABGZODK2EMPRB65444474VRHR4/action/author_attestation","sign_citation":"https://pith.science/pith/ABGZODK2EMPRB65444474VRHR4/action/citation_signature","submit_replication":"https://pith.science/pith/ABGZODK2EMPRB65444474VRHR4/action/replication_record"}},"created_at":"2026-07-05T11:19:03.867512+00:00","updated_at":"2026-07-05T11:19:03.867512+00:00"}