{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:7ZC7IJNIHGFVLDISF36CNNAL53","short_pith_number":"pith:7ZC7IJNI","schema_version":"1.0","canonical_sha256":"fe45f425a8398b558d122efc26b40beef58b72bba7a391b06b449c8b868e3d65","source":{"kind":"arxiv","id":"2102.13019","version":3},"attestation_state":"computed","paper":{"title":"Investigating the Limitations of Transformers with Simple Arithmetic Tasks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Jimmy Lin, Rodrigo Nogueira, Zhiying Jiang","submitted_at":"2021-02-25T17:22:53Z","abstract_excerpt":"The ability to perform arithmetic tasks is a remarkable trait of human intelligence and might form a critical component of more complex reasoning tasks. In this work, we investigate if the surface form of a number has any influence on how sequence-to-sequence language models learn simple arithmetic tasks such as addition and subtraction across a wide range of values. We find that how a number is represented in its surface form has a strong influence on the model's accuracy. In particular, the model fails to learn addition of five-digit numbers when using subwords (e.g., \"32\"), and it struggles"},"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":"2102.13019","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-02-25T17:22:53Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"f123a3164d57feb2d615a585ac1a594b3e7c45ee9a58f345fa5594afb65d1ba4","abstract_canon_sha256":"6372669a07430f76ae1dd468a8e78e543ba10d72cb42f7cc3960aec5a3cb0352"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:31:22.931978Z","signature_b64":"kFi5J1O0tGCq/unzO4TtwWXnNvaiwG4114dKErUf+W9wjOP2EnPFIfzim8Bu5EuH5Up+7bx9d4tpSb18q/VTAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fe45f425a8398b558d122efc26b40beef58b72bba7a391b06b449c8b868e3d65","last_reissued_at":"2026-07-05T02:31:22.931448Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:31:22.931448Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Investigating the Limitations of Transformers with Simple Arithmetic Tasks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Jimmy Lin, Rodrigo Nogueira, Zhiying Jiang","submitted_at":"2021-02-25T17:22:53Z","abstract_excerpt":"The ability to perform arithmetic tasks is a remarkable trait of human intelligence and might form a critical component of more complex reasoning tasks. In this work, we investigate if the surface form of a number has any influence on how sequence-to-sequence language models learn simple arithmetic tasks such as addition and subtraction across a wide range of values. We find that how a number is represented in its surface form has a strong influence on the model's accuracy. In particular, the model fails to learn addition of five-digit numbers when using subwords (e.g., \"32\"), and it struggles"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2102.13019","kind":"arxiv","version":3},"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/2102.13019/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":"2102.13019","created_at":"2026-07-05T02:31:22.931515+00:00"},{"alias_kind":"arxiv_version","alias_value":"2102.13019v3","created_at":"2026-07-05T02:31:22.931515+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2102.13019","created_at":"2026-07-05T02:31:22.931515+00:00"},{"alias_kind":"pith_short_12","alias_value":"7ZC7IJNIHGFV","created_at":"2026-07-05T02:31:22.931515+00:00"},{"alias_kind":"pith_short_16","alias_value":"7ZC7IJNIHGFVLDIS","created_at":"2026-07-05T02:31:22.931515+00:00"},{"alias_kind":"pith_short_8","alias_value":"7ZC7IJNI","created_at":"2026-07-05T02:31:22.931515+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.01391","citing_title":"How Should Transformers Encode Numeric Values in Electronic Health Records?","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02384","citing_title":"TabPrep: Closing the Feature Engineering Gap in Tabular Benchmarks","ref_index":71,"is_internal_anchor":false},{"citing_arxiv_id":"2502.09741","citing_title":"FoNE: Precise Single-Token Number Embeddings via Fourier Features","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2502.12717","citing_title":"Learning the symmetric group: large from small","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2510.06824","citing_title":"Efficient numeracy in language models through single-token number embeddings","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.16675","citing_title":"LinAlg-Bench: A Forensic Benchmark Revealing Structural Failure Modes in LLM Mathematical Reasoning","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2510.18196","citing_title":"Contrastive Decoding Mitigates Score Range Bias in LLM-as-a-Judge","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2402.17762","citing_title":"Massive Activations in Large Language Models","ref_index":66,"is_internal_anchor":false},{"citing_arxiv_id":"2211.10435","citing_title":"PAL: Program-aided Language Models","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13082","citing_title":"The Long Delay to Arithmetic Generalization: When Learned Representations Outrun Behavior","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2108.12409","citing_title":"Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11101","citing_title":"Generating Hadamard matrices with transformers","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11101","citing_title":"Generating Hadamard matrices with transformers","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13242","citing_title":"On the Creativity of AI Agents","ref_index":72,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7ZC7IJNIHGFVLDISF36CNNAL53","json":"https://pith.science/pith/7ZC7IJNIHGFVLDISF36CNNAL53.json","graph_json":"https://pith.science/api/pith-number/7ZC7IJNIHGFVLDISF36CNNAL53/graph.json","events_json":"https://pith.science/api/pith-number/7ZC7IJNIHGFVLDISF36CNNAL53/events.json","paper":"https://pith.science/paper/7ZC7IJNI"},"agent_actions":{"view_html":"https://pith.science/pith/7ZC7IJNIHGFVLDISF36CNNAL53","download_json":"https://pith.science/pith/7ZC7IJNIHGFVLDISF36CNNAL53.json","view_paper":"https://pith.science/paper/7ZC7IJNI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2102.13019&json=true","fetch_graph":"https://pith.science/api/pith-number/7ZC7IJNIHGFVLDISF36CNNAL53/graph.json","fetch_events":"https://pith.science/api/pith-number/7ZC7IJNIHGFVLDISF36CNNAL53/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7ZC7IJNIHGFVLDISF36CNNAL53/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7ZC7IJNIHGFVLDISF36CNNAL53/action/storage_attestation","attest_author":"https://pith.science/pith/7ZC7IJNIHGFVLDISF36CNNAL53/action/author_attestation","sign_citation":"https://pith.science/pith/7ZC7IJNIHGFVLDISF36CNNAL53/action/citation_signature","submit_replication":"https://pith.science/pith/7ZC7IJNIHGFVLDISF36CNNAL53/action/replication_record"}},"created_at":"2026-07-05T02:31:22.931515+00:00","updated_at":"2026-07-05T02:31:22.931515+00:00"}