{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:V23NLQB2IUKMJOGTMT2CXGZC7T","short_pith_number":"pith:V23NLQB2","schema_version":"1.0","canonical_sha256":"aeb6d5c03a4514c4b8d364f42b9b22fcd8593a793b6e62951cc66fac3e522339","source":{"kind":"arxiv","id":"2603.12529","version":2},"attestation_state":"computed","paper":{"title":"TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Terminator trains a predictor on the first position where a reasoning model outputs its final answer to stop chain-of-thought generation early.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alliot Nagle, Ashok Vardhan Makkuva, Dhia Garbaya, Hyeji Kim, Jakhongir Saydaliev, Michael Gastpar","submitted_at":"2026-03-13T00:07:18Z","abstract_excerpt":"Large Reasoning Models (LRMs) achieve impressive performance on complex reasoning tasks via Chain-of-Thought (CoT) reasoning, which enables them to generate intermediate thinking tokens before arriving at the final answer. However, LRMs often suffer from significant overthinking, spending excessive compute time even after the answer is generated early on. Prior work has identified the existence of an optimal reasoning length such that truncating reasoning at this point significantly shortens CoT outputs with virtually no change in performance. However, determining optimal CoT lengths for pract"},"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":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2603.12529","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-03-13T00:07:18Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"52797ecbfb84a1502d208d6fafe3e84ce2c4d850ce607b856f00aa0804fe2273","abstract_canon_sha256":"ff82cd9548dfacdb1ecaf2701cbd5000abac0a3712b55206031490c2516c80a3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:15.798868Z","signature_b64":"2H8mNNqmyMs8W9unbnuGsC9aviXinNsk52h5PvGby8m5AOb1fArwKVCcjMiz9LRAhXvOduhhHEhcz5RgrbV3Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"aeb6d5c03a4514c4b8d364f42b9b22fcd8593a793b6e62951cc66fac3e522339","last_reissued_at":"2026-05-17T23:39:15.798227Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:15.798227Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Terminator trains a predictor on the first position where a reasoning model outputs its final answer to stop chain-of-thought generation early.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alliot Nagle, Ashok Vardhan Makkuva, Dhia Garbaya, Hyeji Kim, Jakhongir Saydaliev, Michael Gastpar","submitted_at":"2026-03-13T00:07:18Z","abstract_excerpt":"Large Reasoning Models (LRMs) achieve impressive performance on complex reasoning tasks via Chain-of-Thought (CoT) reasoning, which enables them to generate intermediate thinking tokens before arriving at the final answer. However, LRMs often suffer from significant overthinking, spending excessive compute time even after the answer is generated early on. Prior work has identified the existence of an optimal reasoning length such that truncating reasoning at this point significantly shortens CoT outputs with virtually no change in performance. However, determining optimal CoT lengths for pract"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Terminator achieves significant reductions in CoT lengths of 14%-55% on average across four challenging practical datasets: MATH-500, AIME 2025, HumanEval, and GPQA, while outperforming current state-of-the-art methods and reducing inference latency by more than 2x compared to the original LRM.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the first position at which the model emits its final answer is a reliable proxy for the optimal stopping point and that a predictor trained on these positions will not degrade accuracy on unseen examples or new models.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Terminator learns to predict optimal early-exit points in chain-of-thought reasoning by training on the first positions where the model emits its final answer, yielding 14-55% shorter outputs with no accuracy loss.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Terminator trains a predictor on the first position where a reasoning model outputs its final answer to stop chain-of-thought generation early.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"d97eca0ee6cf4b4432c1bf5b85f510473f82f8f63100af02db5de0f9bfa13277"},"source":{"id":"2603.12529","kind":"arxiv","version":2},"verdict":{"id":"dca8455d-a2e1-4286-be24-028b3a750900","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T11:09:51.842018Z","strongest_claim":"Terminator achieves significant reductions in CoT lengths of 14%-55% on average across four challenging practical datasets: MATH-500, AIME 2025, HumanEval, and GPQA, while outperforming current state-of-the-art methods and reducing inference latency by more than 2x compared to the original LRM.","one_line_summary":"Terminator learns to predict optimal early-exit points in chain-of-thought reasoning by training on the first positions where the model emits its final answer, yielding 14-55% shorter outputs with no accuracy loss.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the first position at which the model emits its final answer is a reliable proxy for the optimal stopping point and that a predictor trained on these positions will not degrade accuracy on unseen examples or new models.","pith_extraction_headline":"Terminator trains a predictor on the first position where a reasoning model outputs its final answer to stop chain-of-thought generation early."},"references":{"count":11,"sample":[{"doi":"10.18653/v1/2023.emnlp-main","year":2024,"title":"findings-emnlp.633/","work_id":"ad7fc705-f84b-4896-bfd1-d72ce8fb03b6","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Do thinking tokens help or trap? towards more efficient large reasoning model","work_id":"5c46ef72-c90c-4a57-892b-d0d41c19a029","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1038/s41586-025-09422-z","year":2025,"title":"Training Large Language Models to Reason in a Continuous Latent Space","work_id":"3ddd0fd2-c176-408f-9b58-0666c2707f2d","ref_index":3,"cited_arxiv_id":"2412.06769","is_internal_anchor":true},{"doi":"","year":null,"title":"URL https://openreview.net/forum? id=9YvfRrpmyw. Jung, H. and Kim, K.-J. Discrete prompt compression with reinforcement learning.IEEE Access, 12:72578–72587,","work_id":"1d8cb7de-57ed-4d70-9afa-54309609f37b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1109/access.2024","year":2024,"title":"Recurrence-Based Techniques for Data Driven Fault Diagnosis and Monitoring in Neutral-Point-Clamped Inverters","work_id":"ec8f1b74-70c5-45e0-9c70-d60543a0eeba","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":11,"snapshot_sha256":"a1cbcbcb3331a96c45143d6d39491b7de8ea0f4d24533457c387bbe995b92714","internal_anchors":3},"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":"2603.12529","created_at":"2026-05-17T23:39:15.798346+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.12529v2","created_at":"2026-05-17T23:39:15.798346+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.12529","created_at":"2026-05-17T23:39:15.798346+00:00"},{"alias_kind":"pith_short_12","alias_value":"V23NLQB2IUKM","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"V23NLQB2IUKMJOGT","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"V23NLQB2","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/V23NLQB2IUKMJOGTMT2CXGZC7T","json":"https://pith.science/pith/V23NLQB2IUKMJOGTMT2CXGZC7T.json","graph_json":"https://pith.science/api/pith-number/V23NLQB2IUKMJOGTMT2CXGZC7T/graph.json","events_json":"https://pith.science/api/pith-number/V23NLQB2IUKMJOGTMT2CXGZC7T/events.json","paper":"https://pith.science/paper/V23NLQB2"},"agent_actions":{"view_html":"https://pith.science/pith/V23NLQB2IUKMJOGTMT2CXGZC7T","download_json":"https://pith.science/pith/V23NLQB2IUKMJOGTMT2CXGZC7T.json","view_paper":"https://pith.science/paper/V23NLQB2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.12529&json=true","fetch_graph":"https://pith.science/api/pith-number/V23NLQB2IUKMJOGTMT2CXGZC7T/graph.json","fetch_events":"https://pith.science/api/pith-number/V23NLQB2IUKMJOGTMT2CXGZC7T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/V23NLQB2IUKMJOGTMT2CXGZC7T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/V23NLQB2IUKMJOGTMT2CXGZC7T/action/storage_attestation","attest_author":"https://pith.science/pith/V23NLQB2IUKMJOGTMT2CXGZC7T/action/author_attestation","sign_citation":"https://pith.science/pith/V23NLQB2IUKMJOGTMT2CXGZC7T/action/citation_signature","submit_replication":"https://pith.science/pith/V23NLQB2IUKMJOGTMT2CXGZC7T/action/replication_record"}},"created_at":"2026-05-17T23:39:15.798346+00:00","updated_at":"2026-05-17T23:39:15.798346+00:00"}