{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:A2PD2UJSF4M6YTUN5LJD3UQS3P","short_pith_number":"pith:A2PD2UJS","schema_version":"1.0","canonical_sha256":"069e3d51322f19ec4e8dead23dd212dbd22ce94e857b5bcdbf6bb3fb934d8342","source":{"kind":"arxiv","id":"2402.18551","version":2},"attestation_state":"computed","paper":{"title":"Implicit Optimization Bias of Next-Token Prediction in Linear Models","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Christos Thrampoulidis","submitted_at":"2024-02-28T18:34:53Z","abstract_excerpt":"We initiate an investigation into the optimization properties of next-token prediction (NTP), the dominant training paradigm for modern language models. Specifically, we study the structural properties of the solutions selected by gradient-based optimizers among the many possible minimizers of the NTP objective. By framing NTP as cross-entropy minimization across distinct contexts, each tied with a sparse conditional probability distribution across a finite vocabulary of tokens, we introduce \"NTP-separability conditions\" that enable reaching the data-entropy lower bound. With this setup, and f"},"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":"2402.18551","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/publicdomain/zero/1.0/","primary_cat":"cs.LG","submitted_at":"2024-02-28T18:34:53Z","cross_cats_sorted":["cs.CL","stat.ML"],"title_canon_sha256":"9bcefe5fb0932260034c6372fff718a7b42aabb149be8170a82a8087640e8541","abstract_canon_sha256":"03980d4029903321d10c8de4223a07ae17a53dd6df3c5b835443b5d4e6e9103a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:29:14.003278Z","signature_b64":"Zyc405R3shlefq1Vs7EQcVOwxaVzcZp6wqY214se6HRyKXmsuqwLAOV606FKmiXwUQHxneS7UUm6bo0lwN4qCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"069e3d51322f19ec4e8dead23dd212dbd22ce94e857b5bcdbf6bb3fb934d8342","last_reissued_at":"2026-07-05T09:29:14.002826Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:29:14.002826Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Implicit Optimization Bias of Next-Token Prediction in Linear Models","license":"http://creativecommons.org/publicdomain/zero/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Christos Thrampoulidis","submitted_at":"2024-02-28T18:34:53Z","abstract_excerpt":"We initiate an investigation into the optimization properties of next-token prediction (NTP), the dominant training paradigm for modern language models. Specifically, we study the structural properties of the solutions selected by gradient-based optimizers among the many possible minimizers of the NTP objective. By framing NTP as cross-entropy minimization across distinct contexts, each tied with a sparse conditional probability distribution across a finite vocabulary of tokens, we introduce \"NTP-separability conditions\" that enable reaching the data-entropy lower bound. With this setup, and f"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.18551","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/2402.18551/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":"2402.18551","created_at":"2026-07-05T09:29:14.002882+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.18551v2","created_at":"2026-07-05T09:29:14.002882+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.18551","created_at":"2026-07-05T09:29:14.002882+00:00"},{"alias_kind":"pith_short_12","alias_value":"A2PD2UJSF4M6","created_at":"2026-07-05T09:29:14.002882+00:00"},{"alias_kind":"pith_short_16","alias_value":"A2PD2UJSF4M6YTUN","created_at":"2026-07-05T09:29:14.002882+00:00"},{"alias_kind":"pith_short_8","alias_value":"A2PD2UJS","created_at":"2026-07-05T09:29:14.002882+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.11183","citing_title":"On Next-Token Prediction in LLMs: How End Goals Determine the Consistency of Decoding Algorithms","ref_index":29,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/A2PD2UJSF4M6YTUN5LJD3UQS3P","json":"https://pith.science/pith/A2PD2UJSF4M6YTUN5LJD3UQS3P.json","graph_json":"https://pith.science/api/pith-number/A2PD2UJSF4M6YTUN5LJD3UQS3P/graph.json","events_json":"https://pith.science/api/pith-number/A2PD2UJSF4M6YTUN5LJD3UQS3P/events.json","paper":"https://pith.science/paper/A2PD2UJS"},"agent_actions":{"view_html":"https://pith.science/pith/A2PD2UJSF4M6YTUN5LJD3UQS3P","download_json":"https://pith.science/pith/A2PD2UJSF4M6YTUN5LJD3UQS3P.json","view_paper":"https://pith.science/paper/A2PD2UJS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.18551&json=true","fetch_graph":"https://pith.science/api/pith-number/A2PD2UJSF4M6YTUN5LJD3UQS3P/graph.json","fetch_events":"https://pith.science/api/pith-number/A2PD2UJSF4M6YTUN5LJD3UQS3P/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/A2PD2UJSF4M6YTUN5LJD3UQS3P/action/timestamp_anchor","attest_storage":"https://pith.science/pith/A2PD2UJSF4M6YTUN5LJD3UQS3P/action/storage_attestation","attest_author":"https://pith.science/pith/A2PD2UJSF4M6YTUN5LJD3UQS3P/action/author_attestation","sign_citation":"https://pith.science/pith/A2PD2UJSF4M6YTUN5LJD3UQS3P/action/citation_signature","submit_replication":"https://pith.science/pith/A2PD2UJSF4M6YTUN5LJD3UQS3P/action/replication_record"}},"created_at":"2026-07-05T09:29:14.002882+00:00","updated_at":"2026-07-05T09:29:14.002882+00:00"}