{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:TFFA67HPSJ4QQS22R5PMBMABRY","short_pith_number":"pith:TFFA67HP","schema_version":"1.0","canonical_sha256":"994a0f7cef9279084b5a8f5ec0b0018e03ed27e8214b12f74f145fd1cb0f5134","source":{"kind":"arxiv","id":"2409.01659","version":1},"attestation_state":"computed","paper":{"title":"Interpreting and Improving Large Language Models in Arithmetic Calculation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chaoqun Wan, Jieping Ye, Wei Zhang, Xinmei Tian, Xu Shen, Yiu-ming Cheung, Yonggang Zhang","submitted_at":"2024-09-03T07:01:46Z","abstract_excerpt":"Large language models (LLMs) have demonstrated remarkable potential across numerous applications and have shown an emergent ability to tackle complex reasoning tasks, such as mathematical computations. However, even for the simplest arithmetic calculations, the intrinsic mechanisms behind LLMs remain mysterious, making it challenging to ensure reliability. In this work, we delve into uncovering a specific mechanism by which LLMs execute calculations. Through comprehensive experiments, we find that LLMs frequently involve a small fraction (< 5%) of attention heads, which play a pivotal role in "},"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":"2409.01659","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-09-03T07:01:46Z","cross_cats_sorted":[],"title_canon_sha256":"220b5853ebab6c772125cbe463581b11b17f2ff7b299ae6d22c63af53d4002c5","abstract_canon_sha256":"dc37e728dc78255878f88788d2df475519757a92b4aec522b42f90fd23851580"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:02:32.321718Z","signature_b64":"OaFmxz94QpUSQ0MD4gnVqUFkqXaj4dw0k8IlJBLsawCDMBZdvcELC4H1bbUsSbVptXNjoAbPEqwTwR0bblaxCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"994a0f7cef9279084b5a8f5ec0b0018e03ed27e8214b12f74f145fd1cb0f5134","last_reissued_at":"2026-07-05T09:02:32.321244Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:02:32.321244Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Interpreting and Improving Large Language Models in Arithmetic Calculation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Chaoqun Wan, Jieping Ye, Wei Zhang, Xinmei Tian, Xu Shen, Yiu-ming Cheung, Yonggang Zhang","submitted_at":"2024-09-03T07:01:46Z","abstract_excerpt":"Large language models (LLMs) have demonstrated remarkable potential across numerous applications and have shown an emergent ability to tackle complex reasoning tasks, such as mathematical computations. However, even for the simplest arithmetic calculations, the intrinsic mechanisms behind LLMs remain mysterious, making it challenging to ensure reliability. In this work, we delve into uncovering a specific mechanism by which LLMs execute calculations. Through comprehensive experiments, we find that LLMs frequently involve a small fraction (< 5%) of attention heads, which play a pivotal role in "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.01659","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/2409.01659/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":"2409.01659","created_at":"2026-07-05T09:02:32.321299+00:00"},{"alias_kind":"arxiv_version","alias_value":"2409.01659v1","created_at":"2026-07-05T09:02:32.321299+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.01659","created_at":"2026-07-05T09:02:32.321299+00:00"},{"alias_kind":"pith_short_12","alias_value":"TFFA67HPSJ4Q","created_at":"2026-07-05T09:02:32.321299+00:00"},{"alias_kind":"pith_short_16","alias_value":"TFFA67HPSJ4QQS22","created_at":"2026-07-05T09:02:32.321299+00:00"},{"alias_kind":"pith_short_8","alias_value":"TFFA67HP","created_at":"2026-07-05T09:02:32.321299+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.16693","citing_title":"Are Arithmetic Heuristic Neurons Form-Invariant? A Mechanistic Analysis of Symbols, Text, and Code in LLMs","ref_index":71,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TFFA67HPSJ4QQS22R5PMBMABRY","json":"https://pith.science/pith/TFFA67HPSJ4QQS22R5PMBMABRY.json","graph_json":"https://pith.science/api/pith-number/TFFA67HPSJ4QQS22R5PMBMABRY/graph.json","events_json":"https://pith.science/api/pith-number/TFFA67HPSJ4QQS22R5PMBMABRY/events.json","paper":"https://pith.science/paper/TFFA67HP"},"agent_actions":{"view_html":"https://pith.science/pith/TFFA67HPSJ4QQS22R5PMBMABRY","download_json":"https://pith.science/pith/TFFA67HPSJ4QQS22R5PMBMABRY.json","view_paper":"https://pith.science/paper/TFFA67HP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2409.01659&json=true","fetch_graph":"https://pith.science/api/pith-number/TFFA67HPSJ4QQS22R5PMBMABRY/graph.json","fetch_events":"https://pith.science/api/pith-number/TFFA67HPSJ4QQS22R5PMBMABRY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TFFA67HPSJ4QQS22R5PMBMABRY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TFFA67HPSJ4QQS22R5PMBMABRY/action/storage_attestation","attest_author":"https://pith.science/pith/TFFA67HPSJ4QQS22R5PMBMABRY/action/author_attestation","sign_citation":"https://pith.science/pith/TFFA67HPSJ4QQS22R5PMBMABRY/action/citation_signature","submit_replication":"https://pith.science/pith/TFFA67HPSJ4QQS22R5PMBMABRY/action/replication_record"}},"created_at":"2026-07-05T09:02:32.321299+00:00","updated_at":"2026-07-05T09:02:32.321299+00:00"}