{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:CARRBYKCFCOCXOQQ477Z2J6E5X","short_pith_number":"pith:CARRBYKC","schema_version":"1.0","canonical_sha256":"102310e142289c2bba10e7ff9d27c4edcc5c95ac462de4e6de16db39b0679ba5","source":{"kind":"arxiv","id":"2407.07011","version":3},"attestation_state":"computed","paper":{"title":"Induction Heads as an Essential Mechanism for Pattern Matching in In-context Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ekaterina Shutova, Joy Crosbie","submitted_at":"2024-07-09T16:29:21Z","abstract_excerpt":"Large language models (LLMs) have shown a remarkable ability to learn and perform complex tasks through in-context learning (ICL). However, a comprehensive understanding of its internal mechanisms is still lacking. This paper explores the role of induction heads in a few-shot ICL setting. We analyse two state-of-the-art models, Llama-3-8B and InternLM2-20B on abstract pattern recognition and NLP tasks. Our results show that even a minimal ablation of induction heads leads to ICL performance decreases of up to ~32% for abstract pattern recognition tasks, bringing the performance close to random"},"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":"2407.07011","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-09T16:29:21Z","cross_cats_sorted":[],"title_canon_sha256":"1d5c4c0b539fa957dbc68086981839087713a95dcbf32698d659564dc3d94c6c","abstract_canon_sha256":"fcd64a1156de6bccbb3267393ca527fc3ae03a32e6e302b4b8ad8b257338e30d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:43:00.167892Z","signature_b64":"tRvzMOO2YV6w0r9+HwkLOYG+HJ/a24lxIGxnY+UVFXiJ2bi86GZi6zAPMlqjspcMNSsBdw95zdZnNUvzm8DQCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"102310e142289c2bba10e7ff9d27c4edcc5c95ac462de4e6de16db39b0679ba5","last_reissued_at":"2026-07-05T10:43:00.167398Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:43:00.167398Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Induction Heads as an Essential Mechanism for Pattern Matching in In-context Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ekaterina Shutova, Joy Crosbie","submitted_at":"2024-07-09T16:29:21Z","abstract_excerpt":"Large language models (LLMs) have shown a remarkable ability to learn and perform complex tasks through in-context learning (ICL). However, a comprehensive understanding of its internal mechanisms is still lacking. This paper explores the role of induction heads in a few-shot ICL setting. We analyse two state-of-the-art models, Llama-3-8B and InternLM2-20B on abstract pattern recognition and NLP tasks. Our results show that even a minimal ablation of induction heads leads to ICL performance decreases of up to ~32% for abstract pattern recognition tasks, bringing the performance close to random"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.07011","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/2407.07011/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":"2407.07011","created_at":"2026-07-05T10:43:00.167458+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.07011v3","created_at":"2026-07-05T10:43:00.167458+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.07011","created_at":"2026-07-05T10:43:00.167458+00:00"},{"alias_kind":"pith_short_12","alias_value":"CARRBYKCFCOC","created_at":"2026-07-05T10:43:00.167458+00:00"},{"alias_kind":"pith_short_16","alias_value":"CARRBYKCFCOCXOQQ","created_at":"2026-07-05T10:43:00.167458+00:00"},{"alias_kind":"pith_short_8","alias_value":"CARRBYKC","created_at":"2026-07-05T10:43:00.167458+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2509.24164","citing_title":"Localizing Task Recognition and Task Learning in In-Context Learning via Attention Head Analysis","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12671","citing_title":"All Circuits Lead to Rome: Rethinking Functional Anisotropy in Circuit and Sheaf Discovery for LLMs","ref_index":194,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CARRBYKCFCOCXOQQ477Z2J6E5X","json":"https://pith.science/pith/CARRBYKCFCOCXOQQ477Z2J6E5X.json","graph_json":"https://pith.science/api/pith-number/CARRBYKCFCOCXOQQ477Z2J6E5X/graph.json","events_json":"https://pith.science/api/pith-number/CARRBYKCFCOCXOQQ477Z2J6E5X/events.json","paper":"https://pith.science/paper/CARRBYKC"},"agent_actions":{"view_html":"https://pith.science/pith/CARRBYKCFCOCXOQQ477Z2J6E5X","download_json":"https://pith.science/pith/CARRBYKCFCOCXOQQ477Z2J6E5X.json","view_paper":"https://pith.science/paper/CARRBYKC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.07011&json=true","fetch_graph":"https://pith.science/api/pith-number/CARRBYKCFCOCXOQQ477Z2J6E5X/graph.json","fetch_events":"https://pith.science/api/pith-number/CARRBYKCFCOCXOQQ477Z2J6E5X/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CARRBYKCFCOCXOQQ477Z2J6E5X/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CARRBYKCFCOCXOQQ477Z2J6E5X/action/storage_attestation","attest_author":"https://pith.science/pith/CARRBYKCFCOCXOQQ477Z2J6E5X/action/author_attestation","sign_citation":"https://pith.science/pith/CARRBYKCFCOCXOQQ477Z2J6E5X/action/citation_signature","submit_replication":"https://pith.science/pith/CARRBYKCFCOCXOQQ477Z2J6E5X/action/replication_record"}},"created_at":"2026-07-05T10:43:00.167458+00:00","updated_at":"2026-07-05T10:43:00.167458+00:00"}