{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:C5YFUMHETQJMZVCYIWS42DDN6L","short_pith_number":"pith:C5YFUMHE","schema_version":"1.0","canonical_sha256":"17705a30e49c12ccd45845a5cd0c6df2fea5eabf6a0ceee2f0c11b959d0a87e1","source":{"kind":"arxiv","id":"2508.08827","version":1},"attestation_state":"computed","paper":{"title":"TiMoE: Time-Aware Mixture of Language Experts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongyang Fan, Martin Jaggi, Robin Faro, Tamar Alphaidze","submitted_at":"2025-08-12T10:36:36Z","abstract_excerpt":"Large language models (LLMs) are typically trained on fixed snapshots of the web, which means that their knowledge becomes stale and their predictions risk temporal leakage: relying on information that lies in the future relative to a query. We tackle this problem by pre-training from scratch a set of GPT-style experts on disjoint two-year slices of a 2013-2024 corpus and combining them through TiMoE, a Time-aware Mixture of Language Experts. At inference time, TiMoE masks all experts whose training window ends after the query timestamp and merges the remaining log-probabilities in a shared sp"},"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":"2508.08827","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-08-12T10:36:36Z","cross_cats_sorted":[],"title_canon_sha256":"4e9e6121a001a2c30ea44ab1f189310d4a87d6f63c5031242f2b37eb5fbeb304","abstract_canon_sha256":"83e4c06e9c8c26a5ba269feee09ae15e47200bb28d9e2f37983b72ebf3002bba"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:52:34.905111Z","signature_b64":"w9g6pQ9UHOFkmM9dHKAX64pQyME+t5e0valOiJ/Rw71b77jgq1Jb0tcv1gO5tQC+ImAL2oAPlJJV4f7F1hk5DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"17705a30e49c12ccd45845a5cd0c6df2fea5eabf6a0ceee2f0c11b959d0a87e1","last_reissued_at":"2026-07-05T11:52:34.904611Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:52:34.904611Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TiMoE: Time-Aware Mixture of Language Experts","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dongyang Fan, Martin Jaggi, Robin Faro, Tamar Alphaidze","submitted_at":"2025-08-12T10:36:36Z","abstract_excerpt":"Large language models (LLMs) are typically trained on fixed snapshots of the web, which means that their knowledge becomes stale and their predictions risk temporal leakage: relying on information that lies in the future relative to a query. We tackle this problem by pre-training from scratch a set of GPT-style experts on disjoint two-year slices of a 2013-2024 corpus and combining them through TiMoE, a Time-aware Mixture of Language Experts. At inference time, TiMoE masks all experts whose training window ends after the query timestamp and merges the remaining log-probabilities in a shared sp"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.08827","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/2508.08827/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":"2508.08827","created_at":"2026-07-05T11:52:34.904669+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.08827v1","created_at":"2026-07-05T11:52:34.904669+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.08827","created_at":"2026-07-05T11:52:34.904669+00:00"},{"alias_kind":"pith_short_12","alias_value":"C5YFUMHETQJM","created_at":"2026-07-05T11:52:34.904669+00:00"},{"alias_kind":"pith_short_16","alias_value":"C5YFUMHETQJMZVCY","created_at":"2026-07-05T11:52:34.904669+00:00"},{"alias_kind":"pith_short_8","alias_value":"C5YFUMHE","created_at":"2026-07-05T11:52:34.904669+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.22769","citing_title":"Understanding Data Temporality Impact on Large Language Models Pre-training","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2511.21613","citing_title":"Beyond URLs: Metadata Diversity and Position for Efficient LLM Pretraining","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/C5YFUMHETQJMZVCYIWS42DDN6L","json":"https://pith.science/pith/C5YFUMHETQJMZVCYIWS42DDN6L.json","graph_json":"https://pith.science/api/pith-number/C5YFUMHETQJMZVCYIWS42DDN6L/graph.json","events_json":"https://pith.science/api/pith-number/C5YFUMHETQJMZVCYIWS42DDN6L/events.json","paper":"https://pith.science/paper/C5YFUMHE"},"agent_actions":{"view_html":"https://pith.science/pith/C5YFUMHETQJMZVCYIWS42DDN6L","download_json":"https://pith.science/pith/C5YFUMHETQJMZVCYIWS42DDN6L.json","view_paper":"https://pith.science/paper/C5YFUMHE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.08827&json=true","fetch_graph":"https://pith.science/api/pith-number/C5YFUMHETQJMZVCYIWS42DDN6L/graph.json","fetch_events":"https://pith.science/api/pith-number/C5YFUMHETQJMZVCYIWS42DDN6L/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/C5YFUMHETQJMZVCYIWS42DDN6L/action/timestamp_anchor","attest_storage":"https://pith.science/pith/C5YFUMHETQJMZVCYIWS42DDN6L/action/storage_attestation","attest_author":"https://pith.science/pith/C5YFUMHETQJMZVCYIWS42DDN6L/action/author_attestation","sign_citation":"https://pith.science/pith/C5YFUMHETQJMZVCYIWS42DDN6L/action/citation_signature","submit_replication":"https://pith.science/pith/C5YFUMHETQJMZVCYIWS42DDN6L/action/replication_record"}},"created_at":"2026-07-05T11:52:34.904669+00:00","updated_at":"2026-07-05T11:52:34.904669+00:00"}