{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:W5OVOHSU645XQLBVYQC4FE22V3","short_pith_number":"pith:W5OVOHSU","schema_version":"1.0","canonical_sha256":"b75d571e54f73b782c35c405c2935aaecfd035a268bd08bdfa77c36aacf04939","source":{"kind":"arxiv","id":"2502.11196","version":2},"attestation_state":"computed","paper":{"title":"How Do LLMs Acquire New Knowledge? A Knowledge Circuits Perspective on Continual Pre-Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CV","cs.HC"],"primary_cat":"cs.LG","authors_text":"Huajun Chen, Hui Jin, Jiacheng Sun, Ningyu Zhang, Shumin Deng, Yixin Ou, Yunzhi Yao, Zhenguo Li","submitted_at":"2025-02-16T16:55:43Z","abstract_excerpt":"Despite exceptional capabilities in knowledge-intensive tasks, Large Language Models (LLMs) face a critical gap in understanding how they internalize new knowledge, particularly how to structurally embed acquired knowledge in their neural computations. We address this issue through the lens of knowledge circuit evolution, identifying computational subgraphs that facilitate knowledge storage and processing. Our systematic analysis of circuit evolution throughout continual pre-training reveals several key findings: (1) the acquisition of new knowledge is influenced by its relevance to pre-existi"},"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":"2502.11196","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-02-16T16:55:43Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CV","cs.HC"],"title_canon_sha256":"b7f94b8b3021a86416598ead7a9457a7a670ad120c876ae2701d6c2f34492553","abstract_canon_sha256":"92afc980a663650c2613f354bdbfc137ea2b064981e2db4d6e41036f66cc9574"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:13:20.406996Z","signature_b64":"XNwjQmvLNyKgz1v+uHllHDV+Tofrbmj/OQnzYkkO99JcrCTUt5G7pb/WVH6I+r35nOmGhUxOtLbLxTwfWpA0CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b75d571e54f73b782c35c405c2935aaecfd035a268bd08bdfa77c36aacf04939","last_reissued_at":"2026-07-05T11:13:20.406537Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:13:20.406537Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How Do LLMs Acquire New Knowledge? A Knowledge Circuits Perspective on Continual Pre-Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CV","cs.HC"],"primary_cat":"cs.LG","authors_text":"Huajun Chen, Hui Jin, Jiacheng Sun, Ningyu Zhang, Shumin Deng, Yixin Ou, Yunzhi Yao, Zhenguo Li","submitted_at":"2025-02-16T16:55:43Z","abstract_excerpt":"Despite exceptional capabilities in knowledge-intensive tasks, Large Language Models (LLMs) face a critical gap in understanding how they internalize new knowledge, particularly how to structurally embed acquired knowledge in their neural computations. We address this issue through the lens of knowledge circuit evolution, identifying computational subgraphs that facilitate knowledge storage and processing. Our systematic analysis of circuit evolution throughout continual pre-training reveals several key findings: (1) the acquisition of new knowledge is influenced by its relevance to pre-existi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.11196","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/2502.11196/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":"2502.11196","created_at":"2026-07-05T11:13:20.406597+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.11196v2","created_at":"2026-07-05T11:13:20.406597+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.11196","created_at":"2026-07-05T11:13:20.406597+00:00"},{"alias_kind":"pith_short_12","alias_value":"W5OVOHSU645X","created_at":"2026-07-05T11:13:20.406597+00:00"},{"alias_kind":"pith_short_16","alias_value":"W5OVOHSU645XQLBV","created_at":"2026-07-05T11:13:20.406597+00:00"},{"alias_kind":"pith_short_8","alias_value":"W5OVOHSU","created_at":"2026-07-05T11:13:20.406597+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.10554","citing_title":"Benchmarking Knowledge Editing using Logical Rules","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2509.05291","citing_title":"Crosscoding Through Time: Tracking Emergence & Consolidation Of Linguistic Representations Throughout LLM Pretraining","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08192","citing_title":"Inside-Out: Measuring Generalization in Vision Transformers Through Inner Workings","ref_index":53,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/W5OVOHSU645XQLBVYQC4FE22V3","json":"https://pith.science/pith/W5OVOHSU645XQLBVYQC4FE22V3.json","graph_json":"https://pith.science/api/pith-number/W5OVOHSU645XQLBVYQC4FE22V3/graph.json","events_json":"https://pith.science/api/pith-number/W5OVOHSU645XQLBVYQC4FE22V3/events.json","paper":"https://pith.science/paper/W5OVOHSU"},"agent_actions":{"view_html":"https://pith.science/pith/W5OVOHSU645XQLBVYQC4FE22V3","download_json":"https://pith.science/pith/W5OVOHSU645XQLBVYQC4FE22V3.json","view_paper":"https://pith.science/paper/W5OVOHSU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.11196&json=true","fetch_graph":"https://pith.science/api/pith-number/W5OVOHSU645XQLBVYQC4FE22V3/graph.json","fetch_events":"https://pith.science/api/pith-number/W5OVOHSU645XQLBVYQC4FE22V3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/W5OVOHSU645XQLBVYQC4FE22V3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/W5OVOHSU645XQLBVYQC4FE22V3/action/storage_attestation","attest_author":"https://pith.science/pith/W5OVOHSU645XQLBVYQC4FE22V3/action/author_attestation","sign_citation":"https://pith.science/pith/W5OVOHSU645XQLBVYQC4FE22V3/action/citation_signature","submit_replication":"https://pith.science/pith/W5OVOHSU645XQLBVYQC4FE22V3/action/replication_record"}},"created_at":"2026-07-05T11:13:20.406597+00:00","updated_at":"2026-07-05T11:13:20.406597+00:00"}