{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7T4EN2C7SVKEFL6S55GBI4DT5G","short_pith_number":"pith:7T4EN2C7","schema_version":"1.0","canonical_sha256":"fcf846e85f955442afd2ef4c147073e9906697a0538c9d0252b2827fba06433e","source":{"kind":"arxiv","id":"2503.07932","version":2},"attestation_state":"computed","paper":{"title":"A Theory of Learning with Autoregressive Chain of Thought","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CC","cs.LG"],"primary_cat":"stat.ML","authors_text":"Adam Block, Gal Vardi, Nathan Srebro, Nirmit Joshi, Surbhi Goel, Theodor Misiakiewicz, Zhiyuan Li","submitted_at":"2025-03-11T00:21:32Z","abstract_excerpt":"For a given base class of sequence-to-next-token generators, we consider learning prompt-to-answer mappings obtained by iterating a fixed, time-invariant generator for multiple steps, thus generating a chain-of-thought, and then taking the final token as the answer. We formalize the learning problems both when the chain-of-thought is observed and when training only on prompt-answer pairs, with the chain-of-thought latent. We analyze the sample and computational complexity both in terms of general properties of the base class (e.g. its VC dimension) and for specific base classes such as linear "},"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":"2503.07932","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2025-03-11T00:21:32Z","cross_cats_sorted":["cs.AI","cs.CC","cs.LG"],"title_canon_sha256":"daf2741356626cbceeeb63706c223cb61dd717d22c8655f312e330e05ffc9404","abstract_canon_sha256":"42a9bd24813c9371f53ebffce80ec22987a8b713b2934a1a54f03ba065900f2a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:51:23.268292Z","signature_b64":"HnPSUOT5tvrBB0AdIAIkWLoc2tTLALRe4PZvZ+KUhAo8wzcrMpcBq174amxkczp5e780eELK9kzI0R916SgeAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fcf846e85f955442afd2ef4c147073e9906697a0538c9d0252b2827fba06433e","last_reissued_at":"2026-07-05T11:51:23.267789Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:51:23.267789Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Theory of Learning with Autoregressive Chain of Thought","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CC","cs.LG"],"primary_cat":"stat.ML","authors_text":"Adam Block, Gal Vardi, Nathan Srebro, Nirmit Joshi, Surbhi Goel, Theodor Misiakiewicz, Zhiyuan Li","submitted_at":"2025-03-11T00:21:32Z","abstract_excerpt":"For a given base class of sequence-to-next-token generators, we consider learning prompt-to-answer mappings obtained by iterating a fixed, time-invariant generator for multiple steps, thus generating a chain-of-thought, and then taking the final token as the answer. We formalize the learning problems both when the chain-of-thought is observed and when training only on prompt-answer pairs, with the chain-of-thought latent. We analyze the sample and computational complexity both in terms of general properties of the base class (e.g. its VC dimension) and for specific base classes such as linear "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.07932","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/2503.07932/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":"2503.07932","created_at":"2026-07-05T11:51:23.267855+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.07932v2","created_at":"2026-07-05T11:51:23.267855+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.07932","created_at":"2026-07-05T11:51:23.267855+00:00"},{"alias_kind":"pith_short_12","alias_value":"7T4EN2C7SVKE","created_at":"2026-07-05T11:51:23.267855+00:00"},{"alias_kind":"pith_short_16","alias_value":"7T4EN2C7SVKEFL6S","created_at":"2026-07-05T11:51:23.267855+00:00"},{"alias_kind":"pith_short_8","alias_value":"7T4EN2C7","created_at":"2026-07-05T11:51:23.267855+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.12013","citing_title":"Sample Complexity of Autoregressive Reasoning: Chain-of-Thought vs. End-to-End","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06819","citing_title":"A Theory of Online Learning with Autoregressive Chain-of-Thought Reasoning","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7T4EN2C7SVKEFL6S55GBI4DT5G","json":"https://pith.science/pith/7T4EN2C7SVKEFL6S55GBI4DT5G.json","graph_json":"https://pith.science/api/pith-number/7T4EN2C7SVKEFL6S55GBI4DT5G/graph.json","events_json":"https://pith.science/api/pith-number/7T4EN2C7SVKEFL6S55GBI4DT5G/events.json","paper":"https://pith.science/paper/7T4EN2C7"},"agent_actions":{"view_html":"https://pith.science/pith/7T4EN2C7SVKEFL6S55GBI4DT5G","download_json":"https://pith.science/pith/7T4EN2C7SVKEFL6S55GBI4DT5G.json","view_paper":"https://pith.science/paper/7T4EN2C7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.07932&json=true","fetch_graph":"https://pith.science/api/pith-number/7T4EN2C7SVKEFL6S55GBI4DT5G/graph.json","fetch_events":"https://pith.science/api/pith-number/7T4EN2C7SVKEFL6S55GBI4DT5G/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7T4EN2C7SVKEFL6S55GBI4DT5G/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7T4EN2C7SVKEFL6S55GBI4DT5G/action/storage_attestation","attest_author":"https://pith.science/pith/7T4EN2C7SVKEFL6S55GBI4DT5G/action/author_attestation","sign_citation":"https://pith.science/pith/7T4EN2C7SVKEFL6S55GBI4DT5G/action/citation_signature","submit_replication":"https://pith.science/pith/7T4EN2C7SVKEFL6S55GBI4DT5G/action/replication_record"}},"created_at":"2026-07-05T11:51:23.267855+00:00","updated_at":"2026-07-05T11:51:23.267855+00:00"}