{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Z2YITTOP2LIC4AKFV7WGKVG7KY","short_pith_number":"pith:Z2YITTOP","schema_version":"1.0","canonical_sha256":"ceb089cdcfd2d02e0145afec6554df560b40e8038cfc4718df92c0efeb2454be","source":{"kind":"arxiv","id":"2501.18585","version":2},"attestation_state":"computed","paper":{"title":"Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dian Yu, Dong Yu, Haitao Mi, Jiahao Xu, Juntao Li, Linfeng Song, Qiuzhi Liu, Rui Wang, Tian Liang, Xingyu Chen, Yue Wang, Zhaopeng Tu, Zhiwei He, Zhuosheng Zhang","submitted_at":"2025-01-30T18:58:18Z","abstract_excerpt":"Large language models (LLMs) such as OpenAI's o1 have demonstrated remarkable abilities in complex reasoning tasks by scaling test-time compute and exhibiting human-like deep thinking. However, we identify a phenomenon we term underthinking, where o1-like LLMs frequently switch between different reasoning thoughts without sufficiently exploring promising paths to reach a correct solution. This behavior leads to inadequate depth of reasoning and decreased performance, particularly on challenging mathematical problems. To systematically analyze this issue, we conduct experiments on three challen"},"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":"2501.18585","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-01-30T18:58:18Z","cross_cats_sorted":[],"title_canon_sha256":"caf9a356e29625c17c88b7da2dd83adb6e13c8de061deb19d78d4341119f2820","abstract_canon_sha256":"5695e05041698349caaf2de67d71e644b50b702a863861136d1d06dc361e02f6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:16:20.012539Z","signature_b64":"Olr2d6dtSCe76pIxgmWPeeRw3gFDqzOtlupPCFSc0FjS41HEvTsnM1w+Q5FduP+8cUDC+lft0DmQJKVzD+tVBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ceb089cdcfd2d02e0145afec6554df560b40e8038cfc4718df92c0efeb2454be","last_reissued_at":"2026-07-05T10:16:20.012046Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:16:20.012046Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dian Yu, Dong Yu, Haitao Mi, Jiahao Xu, Juntao Li, Linfeng Song, Qiuzhi Liu, Rui Wang, Tian Liang, Xingyu Chen, Yue Wang, Zhaopeng Tu, Zhiwei He, Zhuosheng Zhang","submitted_at":"2025-01-30T18:58:18Z","abstract_excerpt":"Large language models (LLMs) such as OpenAI's o1 have demonstrated remarkable abilities in complex reasoning tasks by scaling test-time compute and exhibiting human-like deep thinking. However, we identify a phenomenon we term underthinking, where o1-like LLMs frequently switch between different reasoning thoughts without sufficiently exploring promising paths to reach a correct solution. This behavior leads to inadequate depth of reasoning and decreased performance, particularly on challenging mathematical problems. To systematically analyze this issue, we conduct experiments on three challen"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.18585","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/2501.18585/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":"2501.18585","created_at":"2026-07-05T10:16:20.012101+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.18585v2","created_at":"2026-07-05T10:16:20.012101+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.18585","created_at":"2026-07-05T10:16:20.012101+00:00"},{"alias_kind":"pith_short_12","alias_value":"Z2YITTOP2LIC","created_at":"2026-07-05T10:16:20.012101+00:00"},{"alias_kind":"pith_short_16","alias_value":"Z2YITTOP2LIC4AKF","created_at":"2026-07-05T10:16:20.012101+00:00"},{"alias_kind":"pith_short_8","alias_value":"Z2YITTOP","created_at":"2026-07-05T10:16:20.012101+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17687","citing_title":"SuCo: Sufficiency-guided Continuous Adaptive Reasoning","ref_index":87,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11470","citing_title":"The Periodic Table of LLM Reasoning: A Structured Survey of Reasoning Paradigms, Methods, and Failure Modes","ref_index":248,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00482","citing_title":"Know When to Stop: Segment-Level Credit Assignment for Reducing Overthinking","ref_index":84,"is_internal_anchor":false},{"citing_arxiv_id":"2607.00862","citing_title":"CAT: Confidence-Adaptive Thinking for Efficient Reasoning of Large Reasoning Models","ref_index":29,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31748","citing_title":"Addressing Over-Refusal in LLMs with Competing Rewards","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16385","citing_title":"Hilbert-Geo: Solving Solid Geometric Problems by Neural-Symbolic Reasoning","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2605.28277","citing_title":"Do LLMs Build World Models From Text? A Multilingual Diagnostic of Spatial Reasoning","ref_index":58,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23074","citing_title":"PathCal: State-Aware Reflection-Marker Calibration for Efficient Reasoning","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2504.02181","citing_title":"A Survey of Scaling in Large Language Model Reasoning","ref_index":213,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16385","citing_title":"Hilbert-Geo: Solving Solid Geometric Problems by Neural-Symbolic Reasoning","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2503.04697","citing_title":"L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2504.01943","citing_title":"OpenCodeReasoning: Advancing Data Distillation for Competitive Coding","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13165","citing_title":"STOP: Structured On-Policy Pruning of Long-Form Reasoning in Low-Data Regimes","ref_index":70,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05715","citing_title":"Decodable but Not Corrected by Fixed Residual-Stream Linear Steering: Evidence from Medical LLM Failure Regimes","ref_index":70,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Z2YITTOP2LIC4AKFV7WGKVG7KY","json":"https://pith.science/pith/Z2YITTOP2LIC4AKFV7WGKVG7KY.json","graph_json":"https://pith.science/api/pith-number/Z2YITTOP2LIC4AKFV7WGKVG7KY/graph.json","events_json":"https://pith.science/api/pith-number/Z2YITTOP2LIC4AKFV7WGKVG7KY/events.json","paper":"https://pith.science/paper/Z2YITTOP"},"agent_actions":{"view_html":"https://pith.science/pith/Z2YITTOP2LIC4AKFV7WGKVG7KY","download_json":"https://pith.science/pith/Z2YITTOP2LIC4AKFV7WGKVG7KY.json","view_paper":"https://pith.science/paper/Z2YITTOP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.18585&json=true","fetch_graph":"https://pith.science/api/pith-number/Z2YITTOP2LIC4AKFV7WGKVG7KY/graph.json","fetch_events":"https://pith.science/api/pith-number/Z2YITTOP2LIC4AKFV7WGKVG7KY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Z2YITTOP2LIC4AKFV7WGKVG7KY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Z2YITTOP2LIC4AKFV7WGKVG7KY/action/storage_attestation","attest_author":"https://pith.science/pith/Z2YITTOP2LIC4AKFV7WGKVG7KY/action/author_attestation","sign_citation":"https://pith.science/pith/Z2YITTOP2LIC4AKFV7WGKVG7KY/action/citation_signature","submit_replication":"https://pith.science/pith/Z2YITTOP2LIC4AKFV7WGKVG7KY/action/replication_record"}},"created_at":"2026-07-05T10:16:20.012101+00:00","updated_at":"2026-07-05T10:16:20.012101+00:00"}