{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:UVR655QA2GNGZSY5XGOCHH3EOV","short_pith_number":"pith:UVR655QA","schema_version":"1.0","canonical_sha256":"a563eef600d19a6ccb1db99c239f64756b2098a6c246047d90b0d64cd9264164","source":{"kind":"arxiv","id":"2502.01694","version":2},"attestation_state":"computed","paper":{"title":"Metastable Dynamics of Chain-of-Thought Reasoning: Provable Benefits of Search, RL and Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.AI","authors_text":"Denny Wu, Jason Lee, Juno Kim, Taiji Suzuki","submitted_at":"2025-02-02T18:19:14Z","abstract_excerpt":"A key paradigm to improve the reasoning capabilities of large language models (LLMs) is to allocate more inference-time compute to search against a verifier or reward model. This process can then be utilized to refine the pretrained model or distill its reasoning patterns into more efficient models. In this paper, we study inference-time compute by viewing chain-of-thought (CoT) generation as a metastable Markov process: easy reasoning steps (e.g., algebraic manipulations) form densely connected clusters, while hard reasoning steps (e.g., applying a relevant theorem) create sparse, low-probabi"},"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":"2502.01694","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-02T18:19:14Z","cross_cats_sorted":["cs.LG","stat.ML"],"title_canon_sha256":"ee896692eeb94bc8a05bed07bef4f244933dea92060a001a41c73cf227db8d66","abstract_canon_sha256":"ff03a273bd90d01b3209426e75cfda9574f4bfb719bc6f5396b8620168f2e50b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:22:12.774120Z","signature_b64":"cNfxWAuPelbhy8SNq2/o03fojuklSnnB9mGh+DJQDXVMqFYKvukHMEfmPL62KCtGAW1/Ta4lZKu67DOLffHrAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a563eef600d19a6ccb1db99c239f64756b2098a6c246047d90b0d64cd9264164","last_reissued_at":"2026-07-05T10:22:12.773574Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:22:12.773574Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Metastable Dynamics of Chain-of-Thought Reasoning: Provable Benefits of Search, RL and Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.AI","authors_text":"Denny Wu, Jason Lee, Juno Kim, Taiji Suzuki","submitted_at":"2025-02-02T18:19:14Z","abstract_excerpt":"A key paradigm to improve the reasoning capabilities of large language models (LLMs) is to allocate more inference-time compute to search against a verifier or reward model. This process can then be utilized to refine the pretrained model or distill its reasoning patterns into more efficient models. In this paper, we study inference-time compute by viewing chain-of-thought (CoT) generation as a metastable Markov process: easy reasoning steps (e.g., algebraic manipulations) form densely connected clusters, while hard reasoning steps (e.g., applying a relevant theorem) create sparse, low-probabi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.01694","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/2502.01694/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":"2502.01694","created_at":"2026-07-05T10:22:12.773635+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.01694v2","created_at":"2026-07-05T10:22:12.773635+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.01694","created_at":"2026-07-05T10:22:12.773635+00:00"},{"alias_kind":"pith_short_12","alias_value":"UVR655QA2GNG","created_at":"2026-07-05T10:22:12.773635+00:00"},{"alias_kind":"pith_short_16","alias_value":"UVR655QA2GNGZSY5","created_at":"2026-07-05T10:22:12.773635+00:00"},{"alias_kind":"pith_short_8","alias_value":"UVR655QA","created_at":"2026-07-05T10:22:12.773635+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2502.17419","citing_title":"From System 1 to System 2: A Survey of Reasoning Large Language Models","ref_index":300,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10237","citing_title":"The Benefits of Temporal Correlations: SGD Learns k-Juntas from Random Walks Efficiently","ref_index":130,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UVR655QA2GNGZSY5XGOCHH3EOV","json":"https://pith.science/pith/UVR655QA2GNGZSY5XGOCHH3EOV.json","graph_json":"https://pith.science/api/pith-number/UVR655QA2GNGZSY5XGOCHH3EOV/graph.json","events_json":"https://pith.science/api/pith-number/UVR655QA2GNGZSY5XGOCHH3EOV/events.json","paper":"https://pith.science/paper/UVR655QA"},"agent_actions":{"view_html":"https://pith.science/pith/UVR655QA2GNGZSY5XGOCHH3EOV","download_json":"https://pith.science/pith/UVR655QA2GNGZSY5XGOCHH3EOV.json","view_paper":"https://pith.science/paper/UVR655QA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.01694&json=true","fetch_graph":"https://pith.science/api/pith-number/UVR655QA2GNGZSY5XGOCHH3EOV/graph.json","fetch_events":"https://pith.science/api/pith-number/UVR655QA2GNGZSY5XGOCHH3EOV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UVR655QA2GNGZSY5XGOCHH3EOV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UVR655QA2GNGZSY5XGOCHH3EOV/action/storage_attestation","attest_author":"https://pith.science/pith/UVR655QA2GNGZSY5XGOCHH3EOV/action/author_attestation","sign_citation":"https://pith.science/pith/UVR655QA2GNGZSY5XGOCHH3EOV/action/citation_signature","submit_replication":"https://pith.science/pith/UVR655QA2GNGZSY5XGOCHH3EOV/action/replication_record"}},"created_at":"2026-07-05T10:22:12.773635+00:00","updated_at":"2026-07-05T10:22:12.773635+00:00"}