{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YHOCY2TOOFS4RXNGD7TUWEJH26","short_pith_number":"pith:YHOCY2TO","schema_version":"1.0","canonical_sha256":"c1dc2c6a6e7165c8dda61fe74b1127d79eed98882225f44f777d417ce1e243ef","source":{"kind":"arxiv","id":"2602.13466","version":2},"attestation_state":"computed","paper":{"title":"Language Model Memory and Memory Models for Language","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Benjamin L. Badger","submitted_at":"2026-02-13T21:16:10Z","abstract_excerpt":"The ability of machine learning models to store input information in hidden layer vector embeddings, analogous to the concept of `memory', is widely employed but not well characterized. We find that language model embeddings typically contain relatively little input information regardless of data and compute scale during training. In contrast, embeddings from autoencoders trained for input regeneration are capable of nearly perfect memory formation. The substitution of memory embeddings for token sequences leads to substantial computational efficiencies, motivating the introduction of a parall"},"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":"2602.13466","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-13T21:16:10Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"07fe29e8ecd80223620a315620ca50f0cdb4891f6ada79970bab346da645c329","abstract_canon_sha256":"78b755192d711a61a9b7de016a3b8e283e69639995952bfb52938b387ec2d8ca"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T01:05:09.172746Z","signature_b64":"RFVzqrJNJRec4IETWwHeORtTIUZh+klKCMIqFn7NpSYfkD3RCXaMbLog8JjLIUTyCv7V4l7/dLG9eFowXT4kCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c1dc2c6a6e7165c8dda61fe74b1127d79eed98882225f44f777d417ce1e243ef","last_reissued_at":"2026-05-20T01:05:09.171801Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T01:05:09.171801Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Language Model Memory and Memory Models for Language","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Benjamin L. Badger","submitted_at":"2026-02-13T21:16:10Z","abstract_excerpt":"The ability of machine learning models to store input information in hidden layer vector embeddings, analogous to the concept of `memory', is widely employed but not well characterized. We find that language model embeddings typically contain relatively little input information regardless of data and compute scale during training. In contrast, embeddings from autoencoders trained for input regeneration are capable of nearly perfect memory formation. The substitution of memory embeddings for token sequences leads to substantial computational efficiencies, motivating the introduction of a parall"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.13466","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/2602.13466/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":"2602.13466","created_at":"2026-05-20T01:05:09.171951+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.13466v2","created_at":"2026-05-20T01:05:09.171951+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.13466","created_at":"2026-05-20T01:05:09.171951+00:00"},{"alias_kind":"pith_short_12","alias_value":"YHOCY2TOOFS4","created_at":"2026-05-20T01:05:09.171951+00:00"},{"alias_kind":"pith_short_16","alias_value":"YHOCY2TOOFS4RXNG","created_at":"2026-05-20T01:05:09.171951+00:00"},{"alias_kind":"pith_short_8","alias_value":"YHOCY2TO","created_at":"2026-05-20T01:05:09.171951+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.08696","citing_title":"Structured Recurrent Mixers for Massively Parallelized Sequence Generation","ref_index":78,"is_internal_anchor":true},{"citing_arxiv_id":"2605.08696","citing_title":"Structured Recurrent Mixers for Massively Parallelized Sequence Generation","ref_index":78,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YHOCY2TOOFS4RXNGD7TUWEJH26","json":"https://pith.science/pith/YHOCY2TOOFS4RXNGD7TUWEJH26.json","graph_json":"https://pith.science/api/pith-number/YHOCY2TOOFS4RXNGD7TUWEJH26/graph.json","events_json":"https://pith.science/api/pith-number/YHOCY2TOOFS4RXNGD7TUWEJH26/events.json","paper":"https://pith.science/paper/YHOCY2TO"},"agent_actions":{"view_html":"https://pith.science/pith/YHOCY2TOOFS4RXNGD7TUWEJH26","download_json":"https://pith.science/pith/YHOCY2TOOFS4RXNGD7TUWEJH26.json","view_paper":"https://pith.science/paper/YHOCY2TO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.13466&json=true","fetch_graph":"https://pith.science/api/pith-number/YHOCY2TOOFS4RXNGD7TUWEJH26/graph.json","fetch_events":"https://pith.science/api/pith-number/YHOCY2TOOFS4RXNGD7TUWEJH26/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YHOCY2TOOFS4RXNGD7TUWEJH26/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YHOCY2TOOFS4RXNGD7TUWEJH26/action/storage_attestation","attest_author":"https://pith.science/pith/YHOCY2TOOFS4RXNGD7TUWEJH26/action/author_attestation","sign_citation":"https://pith.science/pith/YHOCY2TOOFS4RXNGD7TUWEJH26/action/citation_signature","submit_replication":"https://pith.science/pith/YHOCY2TOOFS4RXNGD7TUWEJH26/action/replication_record"}},"created_at":"2026-05-20T01:05:09.171951+00:00","updated_at":"2026-05-20T01:05:09.171951+00:00"}