{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:IZTHVNI3BCMWL6Z7V6HBT5JKVG","short_pith_number":"pith:IZTHVNI3","schema_version":"1.0","canonical_sha256":"46667ab51b089965fb3faf8e19f52aa9a51666562502c94c763de0cdf88e66ba","source":{"kind":"arxiv","id":"2502.03032","version":3},"attestation_state":"computed","paper":{"title":"Analyze Feature Flow to Enhance Interpretation and Steering in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Daniil Gavrilov, Daniil Laptev, Nikita Balagansky, Yaroslav Aksenov","submitted_at":"2025-02-05T09:39:34Z","abstract_excerpt":"We introduce a new approach to systematically map features discovered by sparse autoencoder across consecutive layers of large language models, extending earlier work that examined inter-layer feature links. By using a data-free cosine similarity technique, we trace how specific features persist, transform, or first appear at each stage. This method yields granular flow graphs of feature evolution, enabling fine-grained interpretability and mechanistic insights into model computations. Crucially, we demonstrate how these cross-layer feature maps facilitate direct steering of model behavior by "},"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":"2502.03032","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-05T09:39:34Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"0f1a454297c8edc48a1235f8aeb13b16bc2f9c4241b67fc4339d495557f36e77","abstract_canon_sha256":"a3443d65e80c6b20daeb485eb7c94d218977e84eee4e94e1b2a145d1be403839"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:42:53.256609Z","signature_b64":"PZZj0sNQw5VkatYmupqdDDJo5Wv5DzPxbp4OzrAB4n31W1g/aJ6jNTSk3cbzhcpShHNBZYWfYsH9p+wQQFPVAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"46667ab51b089965fb3faf8e19f52aa9a51666562502c94c763de0cdf88e66ba","last_reissued_at":"2026-07-05T11:42:53.256107Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:42:53.256107Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Analyze Feature Flow to Enhance Interpretation and Steering in Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Daniil Gavrilov, Daniil Laptev, Nikita Balagansky, Yaroslav Aksenov","submitted_at":"2025-02-05T09:39:34Z","abstract_excerpt":"We introduce a new approach to systematically map features discovered by sparse autoencoder across consecutive layers of large language models, extending earlier work that examined inter-layer feature links. By using a data-free cosine similarity technique, we trace how specific features persist, transform, or first appear at each stage. This method yields granular flow graphs of feature evolution, enabling fine-grained interpretability and mechanistic insights into model computations. Crucially, we demonstrate how these cross-layer feature maps facilitate direct steering of model behavior by "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.03032","kind":"arxiv","version":3},"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/2502.03032/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":"2502.03032","created_at":"2026-07-05T11:42:53.256168+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.03032v3","created_at":"2026-07-05T11:42:53.256168+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.03032","created_at":"2026-07-05T11:42:53.256168+00:00"},{"alias_kind":"pith_short_12","alias_value":"IZTHVNI3BCMW","created_at":"2026-07-05T11:42:53.256168+00:00"},{"alias_kind":"pith_short_16","alias_value":"IZTHVNI3BCMWL6Z7","created_at":"2026-07-05T11:42:53.256168+00:00"},{"alias_kind":"pith_short_8","alias_value":"IZTHVNI3","created_at":"2026-07-05T11:42:53.256168+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.12138","citing_title":"Unstable Features, Reproducible Subspaces: Understanding Seed Dependence in Sparse Autoencoders","ref_index":17,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IZTHVNI3BCMWL6Z7V6HBT5JKVG","json":"https://pith.science/pith/IZTHVNI3BCMWL6Z7V6HBT5JKVG.json","graph_json":"https://pith.science/api/pith-number/IZTHVNI3BCMWL6Z7V6HBT5JKVG/graph.json","events_json":"https://pith.science/api/pith-number/IZTHVNI3BCMWL6Z7V6HBT5JKVG/events.json","paper":"https://pith.science/paper/IZTHVNI3"},"agent_actions":{"view_html":"https://pith.science/pith/IZTHVNI3BCMWL6Z7V6HBT5JKVG","download_json":"https://pith.science/pith/IZTHVNI3BCMWL6Z7V6HBT5JKVG.json","view_paper":"https://pith.science/paper/IZTHVNI3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.03032&json=true","fetch_graph":"https://pith.science/api/pith-number/IZTHVNI3BCMWL6Z7V6HBT5JKVG/graph.json","fetch_events":"https://pith.science/api/pith-number/IZTHVNI3BCMWL6Z7V6HBT5JKVG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IZTHVNI3BCMWL6Z7V6HBT5JKVG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IZTHVNI3BCMWL6Z7V6HBT5JKVG/action/storage_attestation","attest_author":"https://pith.science/pith/IZTHVNI3BCMWL6Z7V6HBT5JKVG/action/author_attestation","sign_citation":"https://pith.science/pith/IZTHVNI3BCMWL6Z7V6HBT5JKVG/action/citation_signature","submit_replication":"https://pith.science/pith/IZTHVNI3BCMWL6Z7V6HBT5JKVG/action/replication_record"}},"created_at":"2026-07-05T11:42:53.256168+00:00","updated_at":"2026-07-05T11:42:53.256168+00:00"}