{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:W6HL2ABBDKS6DFQIHYE3NWSGB5","short_pith_number":"pith:W6HL2ABB","schema_version":"1.0","canonical_sha256":"b78ebd00211aa5e196083e09b6da460f5e5176bb9143b0829fb25d8db5932aab","source":{"kind":"arxiv","id":"2309.06979","version":3},"attestation_state":"computed","paper":{"title":"Auto-Regressive Next-Token Predictors are Universal Learners","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Eran Malach","submitted_at":"2023-09-13T14:15:03Z","abstract_excerpt":"Large language models display remarkable capabilities in logical and mathematical reasoning, allowing them to solve complex tasks. Interestingly, these abilities emerge in networks trained on the simple task of next-token prediction. In this work, we present a theoretical framework for studying auto-regressive next-token predictors. We demonstrate that even simple models such as linear next-token predictors, trained on Chain-of-Thought (CoT) data, can approximate any function efficiently computed by a Turing machine. We introduce a new complexity measure -- length complexity -- which measures "},"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":"2309.06979","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-09-13T14:15:03Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"4d1f92712891b48b223613da7a502dcef447b03611962d843750a11386bd9958","abstract_canon_sha256":"07bf4ce554688eac9226a01b21826640cdb2823f649090eab0ab3461cd4d484b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:49:51.462372Z","signature_b64":"S0LTsq7UZKXejpFklRS60RdaISub6JV6Lf8IwdGwdntv/0c82if1N+opC9QgYgUbiw108q+yDwOtWa0bvAPVCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b78ebd00211aa5e196083e09b6da460f5e5176bb9143b0829fb25d8db5932aab","last_reissued_at":"2026-07-05T08:49:51.461910Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:49:51.461910Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Auto-Regressive Next-Token Predictors are Universal Learners","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Eran Malach","submitted_at":"2023-09-13T14:15:03Z","abstract_excerpt":"Large language models display remarkable capabilities in logical and mathematical reasoning, allowing them to solve complex tasks. Interestingly, these abilities emerge in networks trained on the simple task of next-token prediction. In this work, we present a theoretical framework for studying auto-regressive next-token predictors. We demonstrate that even simple models such as linear next-token predictors, trained on Chain-of-Thought (CoT) data, can approximate any function efficiently computed by a Turing machine. We introduce a new complexity measure -- length complexity -- which measures "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2309.06979","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/2309.06979/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":"2309.06979","created_at":"2026-07-05T08:49:51.461969+00:00"},{"alias_kind":"arxiv_version","alias_value":"2309.06979v3","created_at":"2026-07-05T08:49:51.461969+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2309.06979","created_at":"2026-07-05T08:49:51.461969+00:00"},{"alias_kind":"pith_short_12","alias_value":"W6HL2ABBDKS6","created_at":"2026-07-05T08:49:51.461969+00:00"},{"alias_kind":"pith_short_16","alias_value":"W6HL2ABBDKS6DFQI","created_at":"2026-07-05T08:49:51.461969+00:00"},{"alias_kind":"pith_short_8","alias_value":"W6HL2ABB","created_at":"2026-07-05T08:49:51.461969+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06720","citing_title":"When Does In-Context Search Help? A Sampling-Complexity Theory of Reflection-Driven Reasoning","ref_index":12,"is_internal_anchor":true},{"citing_arxiv_id":"2606.09731","citing_title":"Tight Sample Complexity of Transformers","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22471","citing_title":"Lost in Tokenization: Fundamental Trade-offs in Graph Tokenization for Transformers","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2510.26745","citing_title":"Deep sequence models tend to memorize geometrically; it is unclear why","ref_index":114,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06548","citing_title":"Continuous Latent Diffusion Language Model","ref_index":60,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12013","citing_title":"Sample Complexity of Autoregressive Reasoning: Chain-of-Thought vs. End-to-End","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06819","citing_title":"A Theory of Online Learning with Autoregressive Chain-of-Thought Reasoning","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/W6HL2ABBDKS6DFQIHYE3NWSGB5","json":"https://pith.science/pith/W6HL2ABBDKS6DFQIHYE3NWSGB5.json","graph_json":"https://pith.science/api/pith-number/W6HL2ABBDKS6DFQIHYE3NWSGB5/graph.json","events_json":"https://pith.science/api/pith-number/W6HL2ABBDKS6DFQIHYE3NWSGB5/events.json","paper":"https://pith.science/paper/W6HL2ABB"},"agent_actions":{"view_html":"https://pith.science/pith/W6HL2ABBDKS6DFQIHYE3NWSGB5","download_json":"https://pith.science/pith/W6HL2ABBDKS6DFQIHYE3NWSGB5.json","view_paper":"https://pith.science/paper/W6HL2ABB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2309.06979&json=true","fetch_graph":"https://pith.science/api/pith-number/W6HL2ABBDKS6DFQIHYE3NWSGB5/graph.json","fetch_events":"https://pith.science/api/pith-number/W6HL2ABBDKS6DFQIHYE3NWSGB5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/W6HL2ABBDKS6DFQIHYE3NWSGB5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/W6HL2ABBDKS6DFQIHYE3NWSGB5/action/storage_attestation","attest_author":"https://pith.science/pith/W6HL2ABBDKS6DFQIHYE3NWSGB5/action/author_attestation","sign_citation":"https://pith.science/pith/W6HL2ABBDKS6DFQIHYE3NWSGB5/action/citation_signature","submit_replication":"https://pith.science/pith/W6HL2ABBDKS6DFQIHYE3NWSGB5/action/replication_record"}},"created_at":"2026-07-05T08:49:51.461969+00:00","updated_at":"2026-07-05T08:49:51.461969+00:00"}