{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:FAS4XMBDWB4DJICBPEVNEAJLVU","short_pith_number":"pith:FAS4XMBD","schema_version":"1.0","canonical_sha256":"2825cbb023b07834a041792ad2012bad3a7eafd621e8aa80d90ccb616bcc2d2c","source":{"kind":"arxiv","id":"2412.17626","version":3},"attestation_state":"computed","paper":{"title":"Tracking the Feature Dynamics in LLM Training: A Mechanistic Study","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Hao Wang, Hengguan Huang, Yang Xu, Yi Wang","submitted_at":"2024-12-23T14:58:37Z","abstract_excerpt":"Understanding training dynamics and feature evolution is crucial for the mechanistic interpretability of large language models (LLMs). Although sparse autoencoders (SAEs) have been used to identify features within LLMs, a clear picture of how these features evolve during training remains elusive. In this study, we (1) introduce SAE-Track, a novel method for efficiently obtaining a continual series of SAEs, providing the foundation for a mechanistic study that covers (2) the semantic evolution of features, (3) the underlying processes of feature formation, and (4) the directional drift of featu"},"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":"2412.17626","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-12-23T14:58:37Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"bc8c4ffe772b54e356febb7d26151232511ff109ee51856dd1562150decdd747","abstract_canon_sha256":"36d3e929b59a5481c0d2253e5efb65f9e43ce72a7c1b770e182de9c678843434"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:14:51.776268Z","signature_b64":"JirYYw5xp1ZSxFcHqmBNaWHXrEr0oTlA/xsdXaAbNV0sv+ahV9Vg0nimYrNUdKP4qr2KgPjKSr0j6qgA0tyICg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2825cbb023b07834a041792ad2012bad3a7eafd621e8aa80d90ccb616bcc2d2c","last_reissued_at":"2026-07-05T11:14:51.775769Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:14:51.775769Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Tracking the Feature Dynamics in LLM Training: A Mechanistic Study","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Hao Wang, Hengguan Huang, Yang Xu, Yi Wang","submitted_at":"2024-12-23T14:58:37Z","abstract_excerpt":"Understanding training dynamics and feature evolution is crucial for the mechanistic interpretability of large language models (LLMs). Although sparse autoencoders (SAEs) have been used to identify features within LLMs, a clear picture of how these features evolve during training remains elusive. In this study, we (1) introduce SAE-Track, a novel method for efficiently obtaining a continual series of SAEs, providing the foundation for a mechanistic study that covers (2) the semantic evolution of features, (3) the underlying processes of feature formation, and (4) the directional drift of featu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.17626","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/2412.17626/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":"2412.17626","created_at":"2026-07-05T11:14:51.775829+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.17626v3","created_at":"2026-07-05T11:14:51.775829+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.17626","created_at":"2026-07-05T11:14:51.775829+00:00"},{"alias_kind":"pith_short_12","alias_value":"FAS4XMBDWB4D","created_at":"2026-07-05T11:14:51.775829+00:00"},{"alias_kind":"pith_short_16","alias_value":"FAS4XMBDWB4DJICB","created_at":"2026-07-05T11:14:51.775829+00:00"},{"alias_kind":"pith_short_8","alias_value":"FAS4XMBD","created_at":"2026-07-05T11:14:51.775829+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.18789","citing_title":"Features have life history. And we should care","ref_index":4,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FAS4XMBDWB4DJICBPEVNEAJLVU","json":"https://pith.science/pith/FAS4XMBDWB4DJICBPEVNEAJLVU.json","graph_json":"https://pith.science/api/pith-number/FAS4XMBDWB4DJICBPEVNEAJLVU/graph.json","events_json":"https://pith.science/api/pith-number/FAS4XMBDWB4DJICBPEVNEAJLVU/events.json","paper":"https://pith.science/paper/FAS4XMBD"},"agent_actions":{"view_html":"https://pith.science/pith/FAS4XMBDWB4DJICBPEVNEAJLVU","download_json":"https://pith.science/pith/FAS4XMBDWB4DJICBPEVNEAJLVU.json","view_paper":"https://pith.science/paper/FAS4XMBD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.17626&json=true","fetch_graph":"https://pith.science/api/pith-number/FAS4XMBDWB4DJICBPEVNEAJLVU/graph.json","fetch_events":"https://pith.science/api/pith-number/FAS4XMBDWB4DJICBPEVNEAJLVU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FAS4XMBDWB4DJICBPEVNEAJLVU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FAS4XMBDWB4DJICBPEVNEAJLVU/action/storage_attestation","attest_author":"https://pith.science/pith/FAS4XMBDWB4DJICBPEVNEAJLVU/action/author_attestation","sign_citation":"https://pith.science/pith/FAS4XMBDWB4DJICBPEVNEAJLVU/action/citation_signature","submit_replication":"https://pith.science/pith/FAS4XMBDWB4DJICBPEVNEAJLVU/action/replication_record"}},"created_at":"2026-07-05T11:14:51.775829+00:00","updated_at":"2026-07-05T11:14:51.775829+00:00"}