{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:UR2GHQGHRTWLL5IOM4E5XOSKDJ","short_pith_number":"pith:UR2GHQGH","schema_version":"1.0","canonical_sha256":"a47463c0c78cecb5f50e6709dbba4a1a462462f01283854ad36fa60705dd7514","source":{"kind":"arxiv","id":"2403.06963","version":3},"attestation_state":"computed","paper":{"title":"The pitfalls of next-token prediction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Gregor Bachmann, Vaishnavh Nagarajan","submitted_at":"2024-03-11T17:47:30Z","abstract_excerpt":"Can a mere next-token predictor faithfully model human intelligence? We crystallize this emerging concern and correct popular misconceptions surrounding it, and advocate a simple multi-token objective.\n  As a starting point, we argue that the two often-conflated phases of next-token prediction -- autoregressive inference and teacher-forced training -- must be treated distinctly. The popular criticism that errors can compound during autoregressive inference, crucially assumes that teacher-forcing has learned an accurate next-token predictor. This assumption sidesteps a more deep-rooted problem "},"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":"2403.06963","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-03-11T17:47:30Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"3e4a126aa36643595acb39a19e6c9e7b98061285a3fda46fca62ec09833e39a9","abstract_canon_sha256":"2c7b3e96c42e2417788cb051ddcf05396e7123c684da7533a0355ec767e9df53"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:44:39.126690Z","signature_b64":"7qkPRMrIrdb8tJg+LJLGkGh+KhA4WrwfEb55fa9wu7OUjxb5jTUjfh4WjX/w8M1bH1Bi5Rrbkq08unkBabovCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a47463c0c78cecb5f50e6709dbba4a1a462462f01283854ad36fa60705dd7514","last_reissued_at":"2026-07-05T11:44:39.126196Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:44:39.126196Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The pitfalls of next-token prediction","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Gregor Bachmann, Vaishnavh Nagarajan","submitted_at":"2024-03-11T17:47:30Z","abstract_excerpt":"Can a mere next-token predictor faithfully model human intelligence? We crystallize this emerging concern and correct popular misconceptions surrounding it, and advocate a simple multi-token objective.\n  As a starting point, we argue that the two often-conflated phases of next-token prediction -- autoregressive inference and teacher-forced training -- must be treated distinctly. The popular criticism that errors can compound during autoregressive inference, crucially assumes that teacher-forcing has learned an accurate next-token predictor. This assumption sidesteps a more deep-rooted problem "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.06963","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/2403.06963/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":"2403.06963","created_at":"2026-07-05T11:44:39.126253+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.06963v3","created_at":"2026-07-05T11:44:39.126253+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.06963","created_at":"2026-07-05T11:44:39.126253+00:00"},{"alias_kind":"pith_short_12","alias_value":"UR2GHQGHRTWL","created_at":"2026-07-05T11:44:39.126253+00:00"},{"alias_kind":"pith_short_16","alias_value":"UR2GHQGHRTWLL5IO","created_at":"2026-07-05T11:44:39.126253+00:00"},{"alias_kind":"pith_short_8","alias_value":"UR2GHQGH","created_at":"2026-07-05T11:44:39.126253+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.15394","citing_title":"Representation Without Reward: A JEPA Audit for LLM Fine-Tuning","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2506.13351","citing_title":"Direct Reasoning Optimization: Token-Level Reasoning Reflectivity Meets Rubric Gates for Unverifiable Tasks","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2501.00309","citing_title":"Retrieval-Augmented Generation with Graphs (GraphRAG)","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2602.06019","citing_title":"Multi-Token Prediction via Self-Distillation","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06548","citing_title":"Continuous Latent Diffusion Language Model","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.04499","citing_title":"Pen-Strategist: A Reasoning Framework for Penetration Testing Strategy Formation and Analysis","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05329","citing_title":"Semantic Trimming and Auxiliary Multi-step Prediction for Generative Recommendation","ref_index":3,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UR2GHQGHRTWLL5IOM4E5XOSKDJ","json":"https://pith.science/pith/UR2GHQGHRTWLL5IOM4E5XOSKDJ.json","graph_json":"https://pith.science/api/pith-number/UR2GHQGHRTWLL5IOM4E5XOSKDJ/graph.json","events_json":"https://pith.science/api/pith-number/UR2GHQGHRTWLL5IOM4E5XOSKDJ/events.json","paper":"https://pith.science/paper/UR2GHQGH"},"agent_actions":{"view_html":"https://pith.science/pith/UR2GHQGHRTWLL5IOM4E5XOSKDJ","download_json":"https://pith.science/pith/UR2GHQGHRTWLL5IOM4E5XOSKDJ.json","view_paper":"https://pith.science/paper/UR2GHQGH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.06963&json=true","fetch_graph":"https://pith.science/api/pith-number/UR2GHQGHRTWLL5IOM4E5XOSKDJ/graph.json","fetch_events":"https://pith.science/api/pith-number/UR2GHQGHRTWLL5IOM4E5XOSKDJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UR2GHQGHRTWLL5IOM4E5XOSKDJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UR2GHQGHRTWLL5IOM4E5XOSKDJ/action/storage_attestation","attest_author":"https://pith.science/pith/UR2GHQGHRTWLL5IOM4E5XOSKDJ/action/author_attestation","sign_citation":"https://pith.science/pith/UR2GHQGHRTWLL5IOM4E5XOSKDJ/action/citation_signature","submit_replication":"https://pith.science/pith/UR2GHQGHRTWLL5IOM4E5XOSKDJ/action/replication_record"}},"created_at":"2026-07-05T11:44:39.126253+00:00","updated_at":"2026-07-05T11:44:39.126253+00:00"}