{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:EMM72R4UY7S7M2TLVC5CVL7I7K","short_pith_number":"pith:EMM72R4U","schema_version":"1.0","canonical_sha256":"2319fd4794c7e5f66a6ba8ba2aafe8fab01a2590f0f0c70ac1808e159e5d7674","source":{"kind":"arxiv","id":"2305.14160","version":4},"attestation_state":"computed","paper":{"title":"Label Words are Anchors: An Information Flow Perspective for Understanding In-Context Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Damai Dai, Deli Chen, Fandong Meng, Hao Zhou, Jie Zhou, Lean Wang, Lei Li, Xu Sun","submitted_at":"2023-05-23T15:26:20Z","abstract_excerpt":"In-context learning (ICL) emerges as a promising capability of large language models (LLMs) by providing them with demonstration examples to perform diverse tasks. However, the underlying mechanism of how LLMs learn from the provided context remains under-explored. In this paper, we investigate the working mechanism of ICL through an information flow lens. Our findings reveal that label words in the demonstration examples function as anchors: (1) semantic information aggregates into label word representations during the shallow computation layers' processing; (2) the consolidated information i"},"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":"2305.14160","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-05-23T15:26:20Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"2047592355e6d849ef6b03d2d8c4178fb9f5fb32cc222c6fd8e7647cf4547209","abstract_canon_sha256":"e90a3a2a2a348738492094a5818c04dd257ea8c8ac12251a0274a379b733efa9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:25:42.001373Z","signature_b64":"S+NqkQ+zl0LTKRHMrQvTGxu9doAXJgAfHCVFa1pYBieSXlzRFI6kNQBQQbRwmGLnBFY7KzP7Gh9ejXtgHyeUCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2319fd4794c7e5f66a6ba8ba2aafe8fab01a2590f0f0c70ac1808e159e5d7674","last_reissued_at":"2026-07-05T07:25:42.000888Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:25:42.000888Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Label Words are Anchors: An Information Flow Perspective for Understanding In-Context Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Damai Dai, Deli Chen, Fandong Meng, Hao Zhou, Jie Zhou, Lean Wang, Lei Li, Xu Sun","submitted_at":"2023-05-23T15:26:20Z","abstract_excerpt":"In-context learning (ICL) emerges as a promising capability of large language models (LLMs) by providing them with demonstration examples to perform diverse tasks. However, the underlying mechanism of how LLMs learn from the provided context remains under-explored. In this paper, we investigate the working mechanism of ICL through an information flow lens. Our findings reveal that label words in the demonstration examples function as anchors: (1) semantic information aggregates into label word representations during the shallow computation layers' processing; (2) the consolidated information i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.14160","kind":"arxiv","version":4},"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/2305.14160/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":"2305.14160","created_at":"2026-07-05T07:25:42.000948+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.14160v4","created_at":"2026-07-05T07:25:42.000948+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.14160","created_at":"2026-07-05T07:25:42.000948+00:00"},{"alias_kind":"pith_short_12","alias_value":"EMM72R4UY7S7","created_at":"2026-07-05T07:25:42.000948+00:00"},{"alias_kind":"pith_short_16","alias_value":"EMM72R4UY7S7M2TL","created_at":"2026-07-05T07:25:42.000948+00:00"},{"alias_kind":"pith_short_8","alias_value":"EMM72R4U","created_at":"2026-07-05T07:25:42.000948+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2406.02069","citing_title":"PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13403","citing_title":"Why Multimodal In-Context Learning Lags Behind? Unveiling the Inner Mechanisms and Bottlenecks","ref_index":39,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EMM72R4UY7S7M2TLVC5CVL7I7K","json":"https://pith.science/pith/EMM72R4UY7S7M2TLVC5CVL7I7K.json","graph_json":"https://pith.science/api/pith-number/EMM72R4UY7S7M2TLVC5CVL7I7K/graph.json","events_json":"https://pith.science/api/pith-number/EMM72R4UY7S7M2TLVC5CVL7I7K/events.json","paper":"https://pith.science/paper/EMM72R4U"},"agent_actions":{"view_html":"https://pith.science/pith/EMM72R4UY7S7M2TLVC5CVL7I7K","download_json":"https://pith.science/pith/EMM72R4UY7S7M2TLVC5CVL7I7K.json","view_paper":"https://pith.science/paper/EMM72R4U","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.14160&json=true","fetch_graph":"https://pith.science/api/pith-number/EMM72R4UY7S7M2TLVC5CVL7I7K/graph.json","fetch_events":"https://pith.science/api/pith-number/EMM72R4UY7S7M2TLVC5CVL7I7K/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EMM72R4UY7S7M2TLVC5CVL7I7K/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EMM72R4UY7S7M2TLVC5CVL7I7K/action/storage_attestation","attest_author":"https://pith.science/pith/EMM72R4UY7S7M2TLVC5CVL7I7K/action/author_attestation","sign_citation":"https://pith.science/pith/EMM72R4UY7S7M2TLVC5CVL7I7K/action/citation_signature","submit_replication":"https://pith.science/pith/EMM72R4UY7S7M2TLVC5CVL7I7K/action/replication_record"}},"created_at":"2026-07-05T07:25:42.000948+00:00","updated_at":"2026-07-05T07:25:42.000948+00:00"}