{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SZQWMRTU46WTHTDWAZU4X7KGCB","short_pith_number":"pith:SZQWMRTU","schema_version":"1.0","canonical_sha256":"9661664674e7ad33cc760669cbfd461042e80f16878215a3c8d670d5ab988057","source":{"kind":"arxiv","id":"2503.21801","version":1},"attestation_state":"computed","paper":{"title":"Efficient Joint Prediction of Multiple Future Tokens","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alex Lamb, John Langford, Kwangjun Ahn","submitted_at":"2025-03-24T19:52:42Z","abstract_excerpt":"In this short report, we introduce joint multi-token prediction (JTP), a lightweight modification of standard next-token prediction designed to enrich hidden state representations by jointly predicting multiple future tokens. Unlike previous multi-token prediction approaches, JTP strategically employs teacher forcing of future-tokens through a carefully designed representation bottleneck, allowing the model to encode rich predictive information with minimal computational overhead during training. We show that the JTP approach achieves a short-horizon belief state representation, while popular "},"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":"2503.21801","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-03-24T19:52:42Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"a301874fbef106c8982599bfaf02287ae45f9fb857abcb339d65b7d06f78cf4c","abstract_canon_sha256":"79a223989ed8992d4716ce24b12014b6806b834b63b2dbc1dabf2ac482cf040a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:40:42.282389Z","signature_b64":"FlaAttiBh0mYW1nUG2kNO4Td9WiUPBWFpQl/MogFwW/CSBhDiFSUM2uTY8Ml+a+XiZrbZwKh2zfOAoEB4Ge8Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9661664674e7ad33cc760669cbfd461042e80f16878215a3c8d670d5ab988057","last_reissued_at":"2026-07-05T10:40:42.281917Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:40:42.281917Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Efficient Joint Prediction of Multiple Future Tokens","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Alex Lamb, John Langford, Kwangjun Ahn","submitted_at":"2025-03-24T19:52:42Z","abstract_excerpt":"In this short report, we introduce joint multi-token prediction (JTP), a lightweight modification of standard next-token prediction designed to enrich hidden state representations by jointly predicting multiple future tokens. Unlike previous multi-token prediction approaches, JTP strategically employs teacher forcing of future-tokens through a carefully designed representation bottleneck, allowing the model to encode rich predictive information with minimal computational overhead during training. We show that the JTP approach achieves a short-horizon belief state representation, while popular "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.21801","kind":"arxiv","version":1},"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/2503.21801/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":"2503.21801","created_at":"2026-07-05T10:40:42.281980+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.21801v1","created_at":"2026-07-05T10:40:42.281980+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.21801","created_at":"2026-07-05T10:40:42.281980+00:00"},{"alias_kind":"pith_short_12","alias_value":"SZQWMRTU46WT","created_at":"2026-07-05T10:40:42.281980+00:00"},{"alias_kind":"pith_short_16","alias_value":"SZQWMRTU46WTHTDW","created_at":"2026-07-05T10:40:42.281980+00:00"},{"alias_kind":"pith_short_8","alias_value":"SZQWMRTU","created_at":"2026-07-05T10:40:42.281980+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2511.05963","citing_title":"Next-Latent Prediction Transformers Learn Compact World Models","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11912","citing_title":"How Transformers Learn to Plan via Multi-Token Prediction","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SZQWMRTU46WTHTDWAZU4X7KGCB","json":"https://pith.science/pith/SZQWMRTU46WTHTDWAZU4X7KGCB.json","graph_json":"https://pith.science/api/pith-number/SZQWMRTU46WTHTDWAZU4X7KGCB/graph.json","events_json":"https://pith.science/api/pith-number/SZQWMRTU46WTHTDWAZU4X7KGCB/events.json","paper":"https://pith.science/paper/SZQWMRTU"},"agent_actions":{"view_html":"https://pith.science/pith/SZQWMRTU46WTHTDWAZU4X7KGCB","download_json":"https://pith.science/pith/SZQWMRTU46WTHTDWAZU4X7KGCB.json","view_paper":"https://pith.science/paper/SZQWMRTU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.21801&json=true","fetch_graph":"https://pith.science/api/pith-number/SZQWMRTU46WTHTDWAZU4X7KGCB/graph.json","fetch_events":"https://pith.science/api/pith-number/SZQWMRTU46WTHTDWAZU4X7KGCB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SZQWMRTU46WTHTDWAZU4X7KGCB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SZQWMRTU46WTHTDWAZU4X7KGCB/action/storage_attestation","attest_author":"https://pith.science/pith/SZQWMRTU46WTHTDWAZU4X7KGCB/action/author_attestation","sign_citation":"https://pith.science/pith/SZQWMRTU46WTHTDWAZU4X7KGCB/action/citation_signature","submit_replication":"https://pith.science/pith/SZQWMRTU46WTHTDWAZU4X7KGCB/action/replication_record"}},"created_at":"2026-07-05T10:40:42.281980+00:00","updated_at":"2026-07-05T10:40:42.281980+00:00"}