{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:M7JED7IX4VETGUHMA2AN4A5WKQ","short_pith_number":"pith:M7JED7IX","schema_version":"1.0","canonical_sha256":"67d241fd17e5493350ec0680de03b654231ad6c4617f6e4e5f6071b6330d625c","source":{"kind":"arxiv","id":"2303.07971","version":1},"attestation_state":"computed","paper":{"title":"A Theory of Emergent In-Context Learning as Implicit Structure Induction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Michael Hahn, Navin Goyal","submitted_at":"2023-03-14T15:24:05Z","abstract_excerpt":"Scaling large language models (LLMs) leads to an emergent capacity to learn in-context from example demonstrations. Despite progress, theoretical understanding of this phenomenon remains limited. We argue that in-context learning relies on recombination of compositional operations found in natural language data. We derive an information-theoretic bound showing how in-context learning abilities arise from generic next-token prediction when the pretraining distribution has sufficient amounts of compositional structure, under linguistically motivated assumptions. A second bound provides a theoret"},"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":"2303.07971","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-03-14T15:24:05Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"c63a22c3ebdcf663cc4be5ed3010f6d1c1f721905bf0aa4e80e8499f86048e42","abstract_canon_sha256":"163f09cbf029116339b548d1ee2ac3d89447c5b5742c8166625a7d361f129549"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:51:13.547657Z","signature_b64":"2VssRosZ+IhWuKXxbUH4x7AbwUxosMwspYioC5yn4tQHBJPavi4p2YmHmeSuyMvPwK1nm20B2BtcaDQTZ64pAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"67d241fd17e5493350ec0680de03b654231ad6c4617f6e4e5f6071b6330d625c","last_reissued_at":"2026-07-05T05:51:13.547125Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:51:13.547125Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Theory of Emergent In-Context Learning as Implicit Structure Induction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Michael Hahn, Navin Goyal","submitted_at":"2023-03-14T15:24:05Z","abstract_excerpt":"Scaling large language models (LLMs) leads to an emergent capacity to learn in-context from example demonstrations. Despite progress, theoretical understanding of this phenomenon remains limited. We argue that in-context learning relies on recombination of compositional operations found in natural language data. We derive an information-theoretic bound showing how in-context learning abilities arise from generic next-token prediction when the pretraining distribution has sufficient amounts of compositional structure, under linguistically motivated assumptions. A second bound provides a theoret"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2303.07971","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/2303.07971/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":"2303.07971","created_at":"2026-07-05T05:51:13.547202+00:00"},{"alias_kind":"arxiv_version","alias_value":"2303.07971v1","created_at":"2026-07-05T05:51:13.547202+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2303.07971","created_at":"2026-07-05T05:51:13.547202+00:00"},{"alias_kind":"pith_short_12","alias_value":"M7JED7IX4VET","created_at":"2026-07-05T05:51:13.547202+00:00"},{"alias_kind":"pith_short_16","alias_value":"M7JED7IX4VETGUHM","created_at":"2026-07-05T05:51:13.547202+00:00"},{"alias_kind":"pith_short_8","alias_value":"M7JED7IX","created_at":"2026-07-05T05:51:13.547202+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.13607","citing_title":"Reasoning as Pattern Matching: Shared Mechanisms in Human and LLM Everyday Reasoning","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01590","citing_title":"Hawk: Harnessing Hardware-Aware Knowledge for High-Performance NPU Kernel Generation","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2606.00183","citing_title":"Agentic Transformers Provably Learn to Search via Reinforcement Learning","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2511.07372","citing_title":"Provable Benefit of Curriculum in Transformer Tree-Reasoning Post-Training","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2602.07794","citing_title":"Emergent Structured Representations Support Flexible In-Context Inference in Large Language Models","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2301.00234","citing_title":"A Survey on In-context Learning","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05115","citing_title":"Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior","ref_index":295,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/M7JED7IX4VETGUHMA2AN4A5WKQ","json":"https://pith.science/pith/M7JED7IX4VETGUHMA2AN4A5WKQ.json","graph_json":"https://pith.science/api/pith-number/M7JED7IX4VETGUHMA2AN4A5WKQ/graph.json","events_json":"https://pith.science/api/pith-number/M7JED7IX4VETGUHMA2AN4A5WKQ/events.json","paper":"https://pith.science/paper/M7JED7IX"},"agent_actions":{"view_html":"https://pith.science/pith/M7JED7IX4VETGUHMA2AN4A5WKQ","download_json":"https://pith.science/pith/M7JED7IX4VETGUHMA2AN4A5WKQ.json","view_paper":"https://pith.science/paper/M7JED7IX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2303.07971&json=true","fetch_graph":"https://pith.science/api/pith-number/M7JED7IX4VETGUHMA2AN4A5WKQ/graph.json","fetch_events":"https://pith.science/api/pith-number/M7JED7IX4VETGUHMA2AN4A5WKQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/M7JED7IX4VETGUHMA2AN4A5WKQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/M7JED7IX4VETGUHMA2AN4A5WKQ/action/storage_attestation","attest_author":"https://pith.science/pith/M7JED7IX4VETGUHMA2AN4A5WKQ/action/author_attestation","sign_citation":"https://pith.science/pith/M7JED7IX4VETGUHMA2AN4A5WKQ/action/citation_signature","submit_replication":"https://pith.science/pith/M7JED7IX4VETGUHMA2AN4A5WKQ/action/replication_record"}},"created_at":"2026-07-05T05:51:13.547202+00:00","updated_at":"2026-07-05T05:51:13.547202+00:00"}