{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:GE33OONA4SSWAHI5BCCD2AI7CP","short_pith_number":"pith:GE33OONA","schema_version":"1.0","canonical_sha256":"3137b739a0e4a5601d1d08843d011f13e8517785ee45119eee5f9309c3ff5e83","source":{"kind":"arxiv","id":"2306.00802","version":2},"attestation_state":"computed","paper":{"title":"Birth of a Transformer: A Memory Viewpoint","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"stat.ML","authors_text":"Alberto Bietti, Diane Bouchacourt, Herve Jegou, Leon Bottou, Vivien Cabannes","submitted_at":"2023-06-01T15:30:33Z","abstract_excerpt":"Large language models based on transformers have achieved great empirical successes. However, as they are deployed more widely, there is a growing need to better understand their internal mechanisms in order to make them more reliable. These models appear to store vast amounts of knowledge from their training data, and to adapt quickly to new information provided in their context or prompt. We study how transformers balance these two types of knowledge by considering a synthetic setup where tokens are generated from either global or context-specific bigram distributions. By a careful empirical"},"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":"2306.00802","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2023-06-01T15:30:33Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"bdcbf34e0b9168870e11c43471fbcca97694fa90e1ba5f9dbeb95c504c3c4853","abstract_canon_sha256":"30ed789eba9c4753e59773c522660fc14998c7acf931bb5c98bac9d6bcf8e4fc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:09:49.086945Z","signature_b64":"WZHk1ua839ktYe8dEvNh918aNmGs91yrBPgoX+Nc995luhNa+jVT1FBFWZemmovN9CzB8gRtYBtStZvuNPOIAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3137b739a0e4a5601d1d08843d011f13e8517785ee45119eee5f9309c3ff5e83","last_reissued_at":"2026-07-05T07:09:49.086444Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:09:49.086444Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Birth of a Transformer: A Memory Viewpoint","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"stat.ML","authors_text":"Alberto Bietti, Diane Bouchacourt, Herve Jegou, Leon Bottou, Vivien Cabannes","submitted_at":"2023-06-01T15:30:33Z","abstract_excerpt":"Large language models based on transformers have achieved great empirical successes. However, as they are deployed more widely, there is a growing need to better understand their internal mechanisms in order to make them more reliable. These models appear to store vast amounts of knowledge from their training data, and to adapt quickly to new information provided in their context or prompt. We study how transformers balance these two types of knowledge by considering a synthetic setup where tokens are generated from either global or context-specific bigram distributions. By a careful empirical"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2306.00802","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/2306.00802/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":"2306.00802","created_at":"2026-07-05T07:09:49.086502+00:00"},{"alias_kind":"arxiv_version","alias_value":"2306.00802v2","created_at":"2026-07-05T07:09:49.086502+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2306.00802","created_at":"2026-07-05T07:09:49.086502+00:00"},{"alias_kind":"pith_short_12","alias_value":"GE33OONA4SSW","created_at":"2026-07-05T07:09:49.086502+00:00"},{"alias_kind":"pith_short_16","alias_value":"GE33OONA4SSWAHI5","created_at":"2026-07-05T07:09:49.086502+00:00"},{"alias_kind":"pith_short_8","alias_value":"GE33OONA","created_at":"2026-07-05T07:09:49.086502+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06621","citing_title":"Fingerprint, Not Blueprint: How Positional Schemes Set the Default Spectral Algebra of Attention","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2606.26749","citing_title":"Structure Before Collapse: Transient semantic geometry in next-token prediction","ref_index":101,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18820","citing_title":"Emergence of Frontier Superposition: M\\\"obius attractor and Cascade Supervision","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05189","citing_title":"Sharp Capacity Thresholds in Linear Associative Memory: From Winner-Take-All to Listwise Retrieval","ref_index":4,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GE33OONA4SSWAHI5BCCD2AI7CP","json":"https://pith.science/pith/GE33OONA4SSWAHI5BCCD2AI7CP.json","graph_json":"https://pith.science/api/pith-number/GE33OONA4SSWAHI5BCCD2AI7CP/graph.json","events_json":"https://pith.science/api/pith-number/GE33OONA4SSWAHI5BCCD2AI7CP/events.json","paper":"https://pith.science/paper/GE33OONA"},"agent_actions":{"view_html":"https://pith.science/pith/GE33OONA4SSWAHI5BCCD2AI7CP","download_json":"https://pith.science/pith/GE33OONA4SSWAHI5BCCD2AI7CP.json","view_paper":"https://pith.science/paper/GE33OONA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2306.00802&json=true","fetch_graph":"https://pith.science/api/pith-number/GE33OONA4SSWAHI5BCCD2AI7CP/graph.json","fetch_events":"https://pith.science/api/pith-number/GE33OONA4SSWAHI5BCCD2AI7CP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GE33OONA4SSWAHI5BCCD2AI7CP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GE33OONA4SSWAHI5BCCD2AI7CP/action/storage_attestation","attest_author":"https://pith.science/pith/GE33OONA4SSWAHI5BCCD2AI7CP/action/author_attestation","sign_citation":"https://pith.science/pith/GE33OONA4SSWAHI5BCCD2AI7CP/action/citation_signature","submit_replication":"https://pith.science/pith/GE33OONA4SSWAHI5BCCD2AI7CP/action/replication_record"}},"created_at":"2026-07-05T07:09:49.086502+00:00","updated_at":"2026-07-05T07:09:49.086502+00:00"}