{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:XRII22C4LIQS46R7C43YBRKX2U","short_pith_number":"pith:XRII22C4","schema_version":"1.0","canonical_sha256":"bc508d685c5a212e7a3f173780c557d51844d54298cc46c981406f266fdd5ec0","source":{"kind":"arxiv","id":"2505.18373","version":2},"attestation_state":"computed","paper":{"title":"Next-token pretraining implies in-context learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Adam Shai, Eric A. Alt, Henry R. Bigelow, Paul M. Riechers","submitted_at":"2025-05-23T21:00:18Z","abstract_excerpt":"We argue that in-context learning (ICL) predictably arises from standard self-supervised next-token pretraining, rather than being an exotic emergent property. This work establishes the foundational principles of this emergence by focusing on in-distribution ICL, demonstrating how models necessarily adapt to context when trained on token sequences, especially from non-ergodic sources. Our information-theoretic framework precisely predicts these in-distribution ICL dynamics (i.e., context-dependent loss reduction). We verify this with experiments using synthetic datasets of differing types of c"},"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":"2505.18373","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-23T21:00:18Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"7fb56290732a567412907f077c8b642469c1310ea08a8ffaa082b276860e6b4f","abstract_canon_sha256":"6017356276975938a9cb2698fdb5cb2adc1c123f6802a91b303759e0045fbe9f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:35:56.161086Z","signature_b64":"yU4O6vOPZ5CTE+18HGgboEPNbf9M72sJ4MbFVoTzkURU4u3lF1K3j65oOklO9RY1w3gFMtc75bW9/KAJEAwOCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bc508d685c5a212e7a3f173780c557d51844d54298cc46c981406f266fdd5ec0","last_reissued_at":"2026-07-05T11:35:56.160572Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:35:56.160572Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Next-token pretraining implies in-context learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Adam Shai, Eric A. Alt, Henry R. Bigelow, Paul M. Riechers","submitted_at":"2025-05-23T21:00:18Z","abstract_excerpt":"We argue that in-context learning (ICL) predictably arises from standard self-supervised next-token pretraining, rather than being an exotic emergent property. This work establishes the foundational principles of this emergence by focusing on in-distribution ICL, demonstrating how models necessarily adapt to context when trained on token sequences, especially from non-ergodic sources. Our information-theoretic framework precisely predicts these in-distribution ICL dynamics (i.e., context-dependent loss reduction). We verify this with experiments using synthetic datasets of differing types of c"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.18373","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/2505.18373/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":"2505.18373","created_at":"2026-07-05T11:35:56.160640+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.18373v2","created_at":"2026-07-05T11:35:56.160640+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.18373","created_at":"2026-07-05T11:35:56.160640+00:00"},{"alias_kind":"pith_short_12","alias_value":"XRII22C4LIQS","created_at":"2026-07-05T11:35:56.160640+00:00"},{"alias_kind":"pith_short_16","alias_value":"XRII22C4LIQS46R7","created_at":"2026-07-05T11:35:56.160640+00:00"},{"alias_kind":"pith_short_8","alias_value":"XRII22C4","created_at":"2026-07-05T11:35:56.160640+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18089","citing_title":"From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning","ref_index":70,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20382","citing_title":"Do as I Say, Not as I Do: Instruction-Induction Conflict in LLMs","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2605.20382","citing_title":"Do as I Say, Not as I Do: Instruction-Induction Conflict in LLMs","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XRII22C4LIQS46R7C43YBRKX2U","json":"https://pith.science/pith/XRII22C4LIQS46R7C43YBRKX2U.json","graph_json":"https://pith.science/api/pith-number/XRII22C4LIQS46R7C43YBRKX2U/graph.json","events_json":"https://pith.science/api/pith-number/XRII22C4LIQS46R7C43YBRKX2U/events.json","paper":"https://pith.science/paper/XRII22C4"},"agent_actions":{"view_html":"https://pith.science/pith/XRII22C4LIQS46R7C43YBRKX2U","download_json":"https://pith.science/pith/XRII22C4LIQS46R7C43YBRKX2U.json","view_paper":"https://pith.science/paper/XRII22C4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.18373&json=true","fetch_graph":"https://pith.science/api/pith-number/XRII22C4LIQS46R7C43YBRKX2U/graph.json","fetch_events":"https://pith.science/api/pith-number/XRII22C4LIQS46R7C43YBRKX2U/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XRII22C4LIQS46R7C43YBRKX2U/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XRII22C4LIQS46R7C43YBRKX2U/action/storage_attestation","attest_author":"https://pith.science/pith/XRII22C4LIQS46R7C43YBRKX2U/action/author_attestation","sign_citation":"https://pith.science/pith/XRII22C4LIQS46R7C43YBRKX2U/action/citation_signature","submit_replication":"https://pith.science/pith/XRII22C4LIQS46R7C43YBRKX2U/action/replication_record"}},"created_at":"2026-07-05T11:35:56.160640+00:00","updated_at":"2026-07-05T11:35:56.160640+00:00"}