{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:OJFRJBQUELOD6KEXSX2E6DE622","short_pith_number":"pith:OJFRJBQU","schema_version":"1.0","canonical_sha256":"724b14861422dc3f289795f44f0c9ed6b2a9904d5387d9da89f63dbd564d3e5d","source":{"kind":"arxiv","id":"2505.15634","version":4},"attestation_state":"computed","paper":{"title":"Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Haoyi Xiong, Mengnan Du, Xu Wang, Yuzhe Yang, Zihao Li, Ziyu Yao","submitted_at":"2025-05-21T15:17:59Z","abstract_excerpt":"Large Language Models (LLMs) demonstrate the ability to solve reasoning and mathematical problems using the Chain-of-Thought (CoT) technique. Expanding CoT length, as seen in models such as DeepSeek-R1, significantly enhances this reasoning for complex problems, but requires costly and high-quality long CoT data and fine-tuning. This work, inspired by the deep thinking paradigm of DeepSeek-R1, utilizes a steering technique to enhance the reasoning ability of an LLM without external datasets. Our method first employs Sparse Autoencoders (SAEs) to extract interpretable features from vanilla CoT."},"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":"2505.15634","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-21T15:17:59Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"cb7876d287be8925888cc5c2c23f6bcced5b56e77fb2d832bfe59072b2ad2921","abstract_canon_sha256":"36d60cc0a3a41909b68269cb3bb4799b2ceff3b0492965eff57651e3790d43d8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:35:56.070763Z","signature_b64":"c8wVdLtWlIimtaKEukQzqcqYTUJ+ZPY7tlwS2EYIfY16NYfmZykV8JT39a41upKZqXB1vwklwzwaYn86fGSlAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"724b14861422dc3f289795f44f0c9ed6b2a9904d5387d9da89f63dbd564d3e5d","last_reissued_at":"2026-07-05T11:35:56.070231Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:35:56.070231Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Haoyi Xiong, Mengnan Du, Xu Wang, Yuzhe Yang, Zihao Li, Ziyu Yao","submitted_at":"2025-05-21T15:17:59Z","abstract_excerpt":"Large Language Models (LLMs) demonstrate the ability to solve reasoning and mathematical problems using the Chain-of-Thought (CoT) technique. Expanding CoT length, as seen in models such as DeepSeek-R1, significantly enhances this reasoning for complex problems, but requires costly and high-quality long CoT data and fine-tuning. This work, inspired by the deep thinking paradigm of DeepSeek-R1, utilizes a steering technique to enhance the reasoning ability of an LLM without external datasets. Our method first employs Sparse Autoencoders (SAEs) to extract interpretable features from vanilla CoT."},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.15634","kind":"arxiv","version":4},"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/2505.15634/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":"2505.15634","created_at":"2026-07-05T11:35:56.070291+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.15634v4","created_at":"2026-07-05T11:35:56.070291+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.15634","created_at":"2026-07-05T11:35:56.070291+00:00"},{"alias_kind":"pith_short_12","alias_value":"OJFRJBQUELOD","created_at":"2026-07-05T11:35:56.070291+00:00"},{"alias_kind":"pith_short_16","alias_value":"OJFRJBQUELOD6KEX","created_at":"2026-07-05T11:35:56.070291+00:00"},{"alias_kind":"pith_short_8","alias_value":"OJFRJBQU","created_at":"2026-07-05T11:35:56.070291+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.02234","citing_title":"Purified OPSD: On-Policy Self-Distillation Without Losing How to Think","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06188","citing_title":"The Tell-Tale Norm: $\\ell_2$ Magnitude as a Signal for Reasoning Dynamics in Large Language Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19149","citing_title":"How Do Answer Tokens Read Reasoning Traces? Self-Reading Patterns in Thinking LLMs for Quantitative Reasoning","ref_index":26,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OJFRJBQUELOD6KEXSX2E6DE622","json":"https://pith.science/pith/OJFRJBQUELOD6KEXSX2E6DE622.json","graph_json":"https://pith.science/api/pith-number/OJFRJBQUELOD6KEXSX2E6DE622/graph.json","events_json":"https://pith.science/api/pith-number/OJFRJBQUELOD6KEXSX2E6DE622/events.json","paper":"https://pith.science/paper/OJFRJBQU"},"agent_actions":{"view_html":"https://pith.science/pith/OJFRJBQUELOD6KEXSX2E6DE622","download_json":"https://pith.science/pith/OJFRJBQUELOD6KEXSX2E6DE622.json","view_paper":"https://pith.science/paper/OJFRJBQU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.15634&json=true","fetch_graph":"https://pith.science/api/pith-number/OJFRJBQUELOD6KEXSX2E6DE622/graph.json","fetch_events":"https://pith.science/api/pith-number/OJFRJBQUELOD6KEXSX2E6DE622/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OJFRJBQUELOD6KEXSX2E6DE622/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OJFRJBQUELOD6KEXSX2E6DE622/action/storage_attestation","attest_author":"https://pith.science/pith/OJFRJBQUELOD6KEXSX2E6DE622/action/author_attestation","sign_citation":"https://pith.science/pith/OJFRJBQUELOD6KEXSX2E6DE622/action/citation_signature","submit_replication":"https://pith.science/pith/OJFRJBQUELOD6KEXSX2E6DE622/action/replication_record"}},"created_at":"2026-07-05T11:35:56.070291+00:00","updated_at":"2026-07-05T11:35:56.070291+00:00"}