{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GGVG5AYOX5MKFYOAHYUZ7UPIKH","short_pith_number":"pith:GGVG5AYO","schema_version":"1.0","canonical_sha256":"31aa6e830ebf58a2e1c03e299fd1e851c9076bf456a5b472295319c0142f212d","source":{"kind":"arxiv","id":"2503.16334","version":2},"attestation_state":"computed","paper":{"title":"LLM Braces: Straightening Out LLM Predictions with Relevant Sub-Updates","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Lifu Huang, Ying Shen","submitted_at":"2025-03-20T16:55:26Z","abstract_excerpt":"Recent findings reveal that much of the knowledge in a Transformer-based Large Language Model (LLM) is encoded in its feed-forward (FFN) layers, where each FNN layer can be interpreted as the summation of sub-updates, each corresponding to a weighted column vector from the FFN's value parameter matrix that often encodes human-interpretable concepts. In light of this, we hypothesize that model performance and behaviors can be further enhanced and controlled by modulating the contributions of these sub-updates based on their relevance to the input or target output style, and propose LLMBRACES, a"},"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":"2503.16334","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-03-20T16:55:26Z","cross_cats_sorted":[],"title_canon_sha256":"f52473c071c040e79ce10fa02e6c102da09c8549a77dfe4edfd94487af068f2d","abstract_canon_sha256":"e05db826efb89256790958f46117a5358fbefbe7b259347b64d5f99516bf385d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:29:09.315026Z","signature_b64":"lIsMC+k9/OEsultYNPbw0+UR8hiXldwhLeoSh+Zqn4JJr2Dz5RNQsXiakN3tnN+Xe0s6fzskXbE7ehKO0AXZCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"31aa6e830ebf58a2e1c03e299fd1e851c9076bf456a5b472295319c0142f212d","last_reissued_at":"2026-07-05T11:29:09.314480Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:29:09.314480Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LLM Braces: Straightening Out LLM Predictions with Relevant Sub-Updates","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Lifu Huang, Ying Shen","submitted_at":"2025-03-20T16:55:26Z","abstract_excerpt":"Recent findings reveal that much of the knowledge in a Transformer-based Large Language Model (LLM) is encoded in its feed-forward (FFN) layers, where each FNN layer can be interpreted as the summation of sub-updates, each corresponding to a weighted column vector from the FFN's value parameter matrix that often encodes human-interpretable concepts. In light of this, we hypothesize that model performance and behaviors can be further enhanced and controlled by modulating the contributions of these sub-updates based on their relevance to the input or target output style, and propose LLMBRACES, a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.16334","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/2503.16334/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":"2503.16334","created_at":"2026-07-05T11:29:09.314541+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.16334v2","created_at":"2026-07-05T11:29:09.314541+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.16334","created_at":"2026-07-05T11:29:09.314541+00:00"},{"alias_kind":"pith_short_12","alias_value":"GGVG5AYOX5MK","created_at":"2026-07-05T11:29:09.314541+00:00"},{"alias_kind":"pith_short_16","alias_value":"GGVG5AYOX5MKFYOA","created_at":"2026-07-05T11:29:09.314541+00:00"},{"alias_kind":"pith_short_8","alias_value":"GGVG5AYO","created_at":"2026-07-05T11:29:09.314541+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.17153","citing_title":"Amplify Adjacent Token Differences: Enhancing Long Chain-of-Thought Reasoning with Shift-FFN","ref_index":27,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GGVG5AYOX5MKFYOAHYUZ7UPIKH","json":"https://pith.science/pith/GGVG5AYOX5MKFYOAHYUZ7UPIKH.json","graph_json":"https://pith.science/api/pith-number/GGVG5AYOX5MKFYOAHYUZ7UPIKH/graph.json","events_json":"https://pith.science/api/pith-number/GGVG5AYOX5MKFYOAHYUZ7UPIKH/events.json","paper":"https://pith.science/paper/GGVG5AYO"},"agent_actions":{"view_html":"https://pith.science/pith/GGVG5AYOX5MKFYOAHYUZ7UPIKH","download_json":"https://pith.science/pith/GGVG5AYOX5MKFYOAHYUZ7UPIKH.json","view_paper":"https://pith.science/paper/GGVG5AYO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.16334&json=true","fetch_graph":"https://pith.science/api/pith-number/GGVG5AYOX5MKFYOAHYUZ7UPIKH/graph.json","fetch_events":"https://pith.science/api/pith-number/GGVG5AYOX5MKFYOAHYUZ7UPIKH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GGVG5AYOX5MKFYOAHYUZ7UPIKH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GGVG5AYOX5MKFYOAHYUZ7UPIKH/action/storage_attestation","attest_author":"https://pith.science/pith/GGVG5AYOX5MKFYOAHYUZ7UPIKH/action/author_attestation","sign_citation":"https://pith.science/pith/GGVG5AYOX5MKFYOAHYUZ7UPIKH/action/citation_signature","submit_replication":"https://pith.science/pith/GGVG5AYOX5MKFYOAHYUZ7UPIKH/action/replication_record"}},"created_at":"2026-07-05T11:29:09.314541+00:00","updated_at":"2026-07-05T11:29:09.314541+00:00"}