{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:X2LRC2KI2NBWZ4ANMAAXVDXOHE","short_pith_number":"pith:X2LRC2KI","schema_version":"1.0","canonical_sha256":"be97116948d3436cf00d60017a8eee3916a9d2981cf00472ec80f74a93d50055","source":{"kind":"arxiv","id":"2305.15054","version":2},"attestation_state":"computed","paper":{"title":"A Mechanistic Interpretation of Arithmetic Reasoning in Language Models using Causal Mediation Analysis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Alessandro Stolfo, Mrinmaya Sachan, Yonatan Belinkov","submitted_at":"2023-05-24T11:43:47Z","abstract_excerpt":"Mathematical reasoning in large language models (LMs) has garnered significant attention in recent work, but there is a limited understanding of how these models process and store information related to arithmetic tasks within their architecture. In order to improve our understanding of this aspect of language models, we present a mechanistic interpretation of Transformer-based LMs on arithmetic questions using a causal mediation analysis framework. By intervening on the activations of specific model components and measuring the resulting changes in predicted probabilities, we identify the sub"},"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":"2305.15054","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-05-24T11:43:47Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"2c2b103944a3c8bd5a7ebb87c6dd1bae45236bd2e07128854723d6d9634dcf17","abstract_canon_sha256":"f709895f2670aa1c32d1b163000d45c1b28b12ed841e24ab4692f7cd5223b55b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:02:50.804273Z","signature_b64":"1+BBSGfMMUYx1plk3qRuRKNjyUZw3R7MJmu6P9safHnHnO/ipzTB6/C78/v56V392mbeeZjWFvBThlbTy8YOBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"be97116948d3436cf00d60017a8eee3916a9d2981cf00472ec80f74a93d50055","last_reissued_at":"2026-07-05T07:02:50.803807Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:02:50.803807Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Mechanistic Interpretation of Arithmetic Reasoning in Language Models using Causal Mediation Analysis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Alessandro Stolfo, Mrinmaya Sachan, Yonatan Belinkov","submitted_at":"2023-05-24T11:43:47Z","abstract_excerpt":"Mathematical reasoning in large language models (LMs) has garnered significant attention in recent work, but there is a limited understanding of how these models process and store information related to arithmetic tasks within their architecture. In order to improve our understanding of this aspect of language models, we present a mechanistic interpretation of Transformer-based LMs on arithmetic questions using a causal mediation analysis framework. By intervening on the activations of specific model components and measuring the resulting changes in predicted probabilities, we identify the sub"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.15054","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/2305.15054/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":"2305.15054","created_at":"2026-07-05T07:02:50.803863+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.15054v2","created_at":"2026-07-05T07:02:50.803863+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.15054","created_at":"2026-07-05T07:02:50.803863+00:00"},{"alias_kind":"pith_short_12","alias_value":"X2LRC2KI2NBW","created_at":"2026-07-05T07:02:50.803863+00:00"},{"alias_kind":"pith_short_16","alias_value":"X2LRC2KI2NBWZ4AN","created_at":"2026-07-05T07:02:50.803863+00:00"},{"alias_kind":"pith_short_8","alias_value":"X2LRC2KI","created_at":"2026-07-05T07:02:50.803863+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09881","citing_title":"Toward Calibrated, Fair, and accurate Deepfake Detection","ref_index":62,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06840","citing_title":"Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03258","citing_title":"The Right Answer, the Wrong Direction: Why Transformers Fail at Counting and How to Fix It","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2511.17699","citing_title":"Understanding Counting Mechanisms in Large Language and Vision-Language Models","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2309.16042","citing_title":"Towards Best Practices of Activation Patching in Language Models: Metrics and Methods","ref_index":112,"is_internal_anchor":false},{"citing_arxiv_id":"2404.15255","citing_title":"How to use and interpret activation patching","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2605.03258","citing_title":"The Right Answer, the Wrong Direction: Why Transformers Fail at Counting and How to Fix It","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15842","citing_title":"Disentangling Mathematical Reasoning in LLMs: A Methodological Investigation of Internal Mechanisms","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/X2LRC2KI2NBWZ4ANMAAXVDXOHE","json":"https://pith.science/pith/X2LRC2KI2NBWZ4ANMAAXVDXOHE.json","graph_json":"https://pith.science/api/pith-number/X2LRC2KI2NBWZ4ANMAAXVDXOHE/graph.json","events_json":"https://pith.science/api/pith-number/X2LRC2KI2NBWZ4ANMAAXVDXOHE/events.json","paper":"https://pith.science/paper/X2LRC2KI"},"agent_actions":{"view_html":"https://pith.science/pith/X2LRC2KI2NBWZ4ANMAAXVDXOHE","download_json":"https://pith.science/pith/X2LRC2KI2NBWZ4ANMAAXVDXOHE.json","view_paper":"https://pith.science/paper/X2LRC2KI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.15054&json=true","fetch_graph":"https://pith.science/api/pith-number/X2LRC2KI2NBWZ4ANMAAXVDXOHE/graph.json","fetch_events":"https://pith.science/api/pith-number/X2LRC2KI2NBWZ4ANMAAXVDXOHE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/X2LRC2KI2NBWZ4ANMAAXVDXOHE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/X2LRC2KI2NBWZ4ANMAAXVDXOHE/action/storage_attestation","attest_author":"https://pith.science/pith/X2LRC2KI2NBWZ4ANMAAXVDXOHE/action/author_attestation","sign_citation":"https://pith.science/pith/X2LRC2KI2NBWZ4ANMAAXVDXOHE/action/citation_signature","submit_replication":"https://pith.science/pith/X2LRC2KI2NBWZ4ANMAAXVDXOHE/action/replication_record"}},"created_at":"2026-07-05T07:02:50.803863+00:00","updated_at":"2026-07-05T07:02:50.803863+00:00"}