{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:KQMDMWAGY7H4KM4APRQCYY67SC","short_pith_number":"pith:KQMDMWAG","schema_version":"1.0","canonical_sha256":"5418365806c7cfc533807c602c63df90b0e727ab12c815d6f21c846d00b367d5","source":{"kind":"arxiv","id":"2502.12131","version":1},"attestation_state":"computed","paper":{"title":"Transformer Dynamics: A neuroscientific approach to interpretability of large language models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Grigori Guitchounts, Jesseba Fernando","submitted_at":"2025-02-17T18:49:40Z","abstract_excerpt":"As artificial intelligence models have exploded in scale and capability, understanding of their internal mechanisms remains a critical challenge. Inspired by the success of dynamical systems approaches in neuroscience, here we propose a novel framework for studying computations in deep learning systems. We focus on the residual stream (RS) in transformer models, conceptualizing it as a dynamical system evolving across layers. We find that activations of individual RS units exhibit strong continuity across layers, despite the RS being a non-privileged basis. Activations in the RS accelerate and"},"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":"2502.12131","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-17T18:49:40Z","cross_cats_sorted":[],"title_canon_sha256":"ef4477b71549a0a8be3dac7b117bde7a9f0e71b27b0e9f930b1804a67d43865c","abstract_canon_sha256":"7a2a2f42b7005dca150eec82449dd4abe6c08ac5fc1e7eba0422ec2bfe4fb36c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:15:39.483990Z","signature_b64":"YlNzov58D9ptTyYHwcbj/J9S92c+OymFrN+UqmUQdIfSEvDpUtA3jstmdB78js39TnORIJ4dS7bmvJHRVtNuAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5418365806c7cfc533807c602c63df90b0e727ab12c815d6f21c846d00b367d5","last_reissued_at":"2026-07-05T10:15:39.483299Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:15:39.483299Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transformer Dynamics: A neuroscientific approach to interpretability of large language models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Grigori Guitchounts, Jesseba Fernando","submitted_at":"2025-02-17T18:49:40Z","abstract_excerpt":"As artificial intelligence models have exploded in scale and capability, understanding of their internal mechanisms remains a critical challenge. Inspired by the success of dynamical systems approaches in neuroscience, here we propose a novel framework for studying computations in deep learning systems. We focus on the residual stream (RS) in transformer models, conceptualizing it as a dynamical system evolving across layers. We find that activations of individual RS units exhibit strong continuity across layers, despite the RS being a non-privileged basis. Activations in the RS accelerate and"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12131","kind":"arxiv","version":1},"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/2502.12131/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":"2502.12131","created_at":"2026-07-05T10:15:39.483379+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.12131v1","created_at":"2026-07-05T10:15:39.483379+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12131","created_at":"2026-07-05T10:15:39.483379+00:00"},{"alias_kind":"pith_short_12","alias_value":"KQMDMWAGY7H4","created_at":"2026-07-05T10:15:39.483379+00:00"},{"alias_kind":"pith_short_16","alias_value":"KQMDMWAGY7H4KM4A","created_at":"2026-07-05T10:15:39.483379+00:00"},{"alias_kind":"pith_short_8","alias_value":"KQMDMWAG","created_at":"2026-07-05T10:15:39.483379+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.14258","citing_title":"Dynamics of the Transformer Residual Stream: Coupling Spectral Geometry to Network Topology","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12016","citing_title":"Identity as Attractor: Geometric Evidence for Persistent Agent Architecture in LLM Activation Space","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KQMDMWAGY7H4KM4APRQCYY67SC","json":"https://pith.science/pith/KQMDMWAGY7H4KM4APRQCYY67SC.json","graph_json":"https://pith.science/api/pith-number/KQMDMWAGY7H4KM4APRQCYY67SC/graph.json","events_json":"https://pith.science/api/pith-number/KQMDMWAGY7H4KM4APRQCYY67SC/events.json","paper":"https://pith.science/paper/KQMDMWAG"},"agent_actions":{"view_html":"https://pith.science/pith/KQMDMWAGY7H4KM4APRQCYY67SC","download_json":"https://pith.science/pith/KQMDMWAGY7H4KM4APRQCYY67SC.json","view_paper":"https://pith.science/paper/KQMDMWAG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.12131&json=true","fetch_graph":"https://pith.science/api/pith-number/KQMDMWAGY7H4KM4APRQCYY67SC/graph.json","fetch_events":"https://pith.science/api/pith-number/KQMDMWAGY7H4KM4APRQCYY67SC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KQMDMWAGY7H4KM4APRQCYY67SC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KQMDMWAGY7H4KM4APRQCYY67SC/action/storage_attestation","attest_author":"https://pith.science/pith/KQMDMWAGY7H4KM4APRQCYY67SC/action/author_attestation","sign_citation":"https://pith.science/pith/KQMDMWAGY7H4KM4APRQCYY67SC/action/citation_signature","submit_replication":"https://pith.science/pith/KQMDMWAGY7H4KM4APRQCYY67SC/action/replication_record"}},"created_at":"2026-07-05T10:15:39.483379+00:00","updated_at":"2026-07-05T10:15:39.483379+00:00"}