{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EQJLXUERYRTECSZVSWKDUZMVDT","short_pith_number":"pith:EQJLXUER","schema_version":"1.0","canonical_sha256":"2412bbd091c466414b3595943a65951cd72e8438ee2ab88f722d86b6a7c9f7b6","source":{"kind":"arxiv","id":"2506.03616","version":1},"attestation_state":"computed","paper":{"title":"Learning to Insert [PAUSE] Tokens for Better Reasoning","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Eunki Kim, James Thorne, Sangryul Kim","submitted_at":"2025-06-04T06:48:41Z","abstract_excerpt":"To enhance reasoning capabilities, previous works have explored incorporating special-purpose tokens into the training process. These strategies strengthen the learning mechanism of transformer-based large language models (LLMs). Building on prior research, in which inserting dummy tokens consecutively just before reasoning steps can enhance effectiveness, we introduce a novel approach termed Dynamic Inserting Tokens Training (DIT). Our method identifies positions within sequences where model confidence is lowest according to token log-likelihood. Strategically inserting [PAUSE] tokens on thes"},"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":"2506.03616","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-04T06:48:41Z","cross_cats_sorted":[],"title_canon_sha256":"f6e369e99dae18c74166178e44f971e61748e737a4c0645c19939eea987584fe","abstract_canon_sha256":"f6befa4fc191694643884e6b7137d5a4fd9e5b454cdd64ebc24358457dbcdd39"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:51:28.076462Z","signature_b64":"sC+J1yt55q5bMbBliy0RxvgNdtcT8dhUK3B6Pg9lt6i7gq/Kh3u1YnsjY1iwzF33DNx0Qyze4jEg1lPTzk5yDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2412bbd091c466414b3595943a65951cd72e8438ee2ab88f722d86b6a7c9f7b6","last_reissued_at":"2026-07-05T11:51:28.075922Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:51:28.075922Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning to Insert [PAUSE] Tokens for Better Reasoning","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Eunki Kim, James Thorne, Sangryul Kim","submitted_at":"2025-06-04T06:48:41Z","abstract_excerpt":"To enhance reasoning capabilities, previous works have explored incorporating special-purpose tokens into the training process. These strategies strengthen the learning mechanism of transformer-based large language models (LLMs). Building on prior research, in which inserting dummy tokens consecutively just before reasoning steps can enhance effectiveness, we introduce a novel approach termed Dynamic Inserting Tokens Training (DIT). Our method identifies positions within sequences where model confidence is lowest according to token log-likelihood. Strategically inserting [PAUSE] tokens on thes"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.03616","kind":"arxiv","version":1},"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/2506.03616/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":"2506.03616","created_at":"2026-07-05T11:51:28.075984+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.03616v1","created_at":"2026-07-05T11:51:28.075984+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.03616","created_at":"2026-07-05T11:51:28.075984+00:00"},{"alias_kind":"pith_short_12","alias_value":"EQJLXUERYRTE","created_at":"2026-07-05T11:51:28.075984+00:00"},{"alias_kind":"pith_short_16","alias_value":"EQJLXUERYRTECSZV","created_at":"2026-07-05T11:51:28.075984+00:00"},{"alias_kind":"pith_short_8","alias_value":"EQJLXUER","created_at":"2026-07-05T11:51:28.075984+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.07021","citing_title":"Behavior Cue Reasoning: Monitorable Reasoning Improves Efficiency and Safety through Oversight","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.07021","citing_title":"Behavior Cue Reasoning: Monitorable Reasoning Improves Efficiency and Safety through Oversight","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EQJLXUERYRTECSZVSWKDUZMVDT","json":"https://pith.science/pith/EQJLXUERYRTECSZVSWKDUZMVDT.json","graph_json":"https://pith.science/api/pith-number/EQJLXUERYRTECSZVSWKDUZMVDT/graph.json","events_json":"https://pith.science/api/pith-number/EQJLXUERYRTECSZVSWKDUZMVDT/events.json","paper":"https://pith.science/paper/EQJLXUER"},"agent_actions":{"view_html":"https://pith.science/pith/EQJLXUERYRTECSZVSWKDUZMVDT","download_json":"https://pith.science/pith/EQJLXUERYRTECSZVSWKDUZMVDT.json","view_paper":"https://pith.science/paper/EQJLXUER","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.03616&json=true","fetch_graph":"https://pith.science/api/pith-number/EQJLXUERYRTECSZVSWKDUZMVDT/graph.json","fetch_events":"https://pith.science/api/pith-number/EQJLXUERYRTECSZVSWKDUZMVDT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EQJLXUERYRTECSZVSWKDUZMVDT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EQJLXUERYRTECSZVSWKDUZMVDT/action/storage_attestation","attest_author":"https://pith.science/pith/EQJLXUERYRTECSZVSWKDUZMVDT/action/author_attestation","sign_citation":"https://pith.science/pith/EQJLXUERYRTECSZVSWKDUZMVDT/action/citation_signature","submit_replication":"https://pith.science/pith/EQJLXUERYRTECSZVSWKDUZMVDT/action/replication_record"}},"created_at":"2026-07-05T11:51:28.075984+00:00","updated_at":"2026-07-05T11:51:28.075984+00:00"}