{"paper":{"title":"Continual Knowledge Updating in LLM Systems: Learning Through Multi-Timescale Memory Dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Coupling fast and slow variables on knowledge-graph edges lets external memory adapt on its own for continual LLM updates.","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Andreas Pattichis, Constantine Dovrolis","submitted_at":"2026-05-06T16:33:42Z","abstract_excerpt":"LLMs are trained once, then deployed into a world that never stops changing. External memory compensates for this, but most systems manage it explicitly rather than letting it adapt on its own. Biological memory works differently: coupled multi-timescale dynamics make new associations immediately usable, strengthen what repetition confirms, and let the rest fade. We argue that external memory should follow a similar principle. In Memini, this view takes the form of an associative memory that organizes knowledge as a directed graph. Each edge carries two coupled internal variables, one fast and"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"From this coupling, episodic sensitivity, gradual consolidation, and selective forgetting emerge as facets of a single mechanism, reframing external memory as a learning substrate that reorganizes through its own dynamics.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the Benna-Fusi multi-timescale coupling, when placed on the edges of an LLM knowledge graph, will produce stable continual learning without introducing interference, scalability bottlenecks, or loss of previously consolidated knowledge.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Memini organizes LLM knowledge as a directed graph whose edges follow coupled fast-slow dynamics so that episodic recall, consolidation, and selective forgetting arise automatically from a single mechanism.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Coupling fast and slow variables on knowledge-graph edges lets external memory adapt on its own for continual LLM updates.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"513fae4c959da520176731ef54035c938b8c17cc2d5e2a7b118cf41214ef6721"},"source":{"id":"2605.05097","kind":"arxiv","version":3},"verdict":{"id":"6fa7af5d-429b-4df2-98b8-e27a0fda45ef","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T18:08:36.065136Z","strongest_claim":"From this coupling, episodic sensitivity, gradual consolidation, and selective forgetting emerge as facets of a single mechanism, reframing external memory as a learning substrate that reorganizes through its own dynamics.","one_line_summary":"Memini organizes LLM knowledge as a directed graph whose edges follow coupled fast-slow dynamics so that episodic recall, consolidation, and selective forgetting arise automatically from a single mechanism.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the Benna-Fusi multi-timescale coupling, when placed on the edges of an LLM knowledge graph, will produce stable continual learning without introducing interference, scalability bottlenecks, or loss of previously consolidated knowledge.","pith_extraction_headline":"Coupling fast and slow variables on knowledge-graph edges lets external memory adapt on its own for continual LLM updates."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.05097/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T10:37:43.284813Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T21:31:19.601930Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T13:50:07.744547Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"2128df2ec109b8f96b036d3228818ddf9f3f8b8f9961d8b32145dee004f9f22a"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":3,"snapshot_sha256":"67c6fae30e57087f56563c9f3d2d10cdc3b61e0a4b08566c38a6aee471437c50"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}