{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VDUVT3T63W225YQ55LTAL2VQ4D","short_pith_number":"pith:VDUVT3T6","schema_version":"1.0","canonical_sha256":"a8e959ee7eddb5aee21deae605eab0e0d10fd6c16872e3ea66b936e6ef63d010","source":{"kind":"arxiv","id":"2501.19406","version":2},"attestation_state":"computed","paper":{"title":"Low-Rank Adapting Models for Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Joshua Engels, Matthew Chen, Max Tegmark","submitted_at":"2025-01-31T18:59:16Z","abstract_excerpt":"Sparse autoencoders (SAEs) decompose language model representations into a sparse set of linear latent vectors. Recent works have improved SAEs using language model gradients, but these techniques require many expensive backward passes during training and still cause a significant increase in cross entropy loss when SAE reconstructions are inserted into the model. In this work, we improve on these limitations by taking a fundamentally different approach: we use low-rank adaptation (LoRA) to finetune the \\textit{language model itself} around a previously trained SAE. We analyze our method acros"},"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":"2501.19406","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-01-31T18:59:16Z","cross_cats_sorted":[],"title_canon_sha256":"827c9cd142f925ef2cafeb40e3e193826aab8eb987f3eeac21e42ed084e7abb9","abstract_canon_sha256":"9ffcdf01a24a34b13d9baf6eae2b82b1930b192ecff69d12de9dda62d391f0e7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:10:04.386230Z","signature_b64":"QOkqp0hVFIuaUhTRpcFHRZ0mfuTvRZgZDIL/poxvqOvi34+lDHZyhXBfsGr6WkoXOGplJAiAnTb04kZQRz5SCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a8e959ee7eddb5aee21deae605eab0e0d10fd6c16872e3ea66b936e6ef63d010","last_reissued_at":"2026-07-05T11:10:04.385699Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:10:04.385699Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Low-Rank Adapting Models for Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Joshua Engels, Matthew Chen, Max Tegmark","submitted_at":"2025-01-31T18:59:16Z","abstract_excerpt":"Sparse autoencoders (SAEs) decompose language model representations into a sparse set of linear latent vectors. Recent works have improved SAEs using language model gradients, but these techniques require many expensive backward passes during training and still cause a significant increase in cross entropy loss when SAE reconstructions are inserted into the model. In this work, we improve on these limitations by taking a fundamentally different approach: we use low-rank adaptation (LoRA) to finetune the \\textit{language model itself} around a previously trained SAE. We analyze our method acros"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.19406","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/2501.19406/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":"2501.19406","created_at":"2026-07-05T11:10:04.385763+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.19406v2","created_at":"2026-07-05T11:10:04.385763+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.19406","created_at":"2026-07-05T11:10:04.385763+00:00"},{"alias_kind":"pith_short_12","alias_value":"VDUVT3T63W22","created_at":"2026-07-05T11:10:04.385763+00:00"},{"alias_kind":"pith_short_16","alias_value":"VDUVT3T63W225YQ5","created_at":"2026-07-05T11:10:04.385763+00:00"},{"alias_kind":"pith_short_8","alias_value":"VDUVT3T6","created_at":"2026-07-05T11:10:04.385763+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.09967","citing_title":"Resa: Transparent Reasoning Models via SAEs","ref_index":5,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VDUVT3T63W225YQ55LTAL2VQ4D","json":"https://pith.science/pith/VDUVT3T63W225YQ55LTAL2VQ4D.json","graph_json":"https://pith.science/api/pith-number/VDUVT3T63W225YQ55LTAL2VQ4D/graph.json","events_json":"https://pith.science/api/pith-number/VDUVT3T63W225YQ55LTAL2VQ4D/events.json","paper":"https://pith.science/paper/VDUVT3T6"},"agent_actions":{"view_html":"https://pith.science/pith/VDUVT3T63W225YQ55LTAL2VQ4D","download_json":"https://pith.science/pith/VDUVT3T63W225YQ55LTAL2VQ4D.json","view_paper":"https://pith.science/paper/VDUVT3T6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.19406&json=true","fetch_graph":"https://pith.science/api/pith-number/VDUVT3T63W225YQ55LTAL2VQ4D/graph.json","fetch_events":"https://pith.science/api/pith-number/VDUVT3T63W225YQ55LTAL2VQ4D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VDUVT3T63W225YQ55LTAL2VQ4D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VDUVT3T63W225YQ55LTAL2VQ4D/action/storage_attestation","attest_author":"https://pith.science/pith/VDUVT3T63W225YQ55LTAL2VQ4D/action/author_attestation","sign_citation":"https://pith.science/pith/VDUVT3T63W225YQ55LTAL2VQ4D/action/citation_signature","submit_replication":"https://pith.science/pith/VDUVT3T63W225YQ55LTAL2VQ4D/action/replication_record"}},"created_at":"2026-07-05T11:10:04.385763+00:00","updated_at":"2026-07-05T11:10:04.385763+00:00"}