{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:MBJ6M2H7MXBUQQGMY72JF5MV4V","short_pith_number":"pith:MBJ6M2H7","schema_version":"1.0","canonical_sha256":"6053e668ff65c34840ccc7f492f595e57033e5f2292ff90555ebe31500c44c01","source":{"kind":"arxiv","id":"2507.12990","version":1},"attestation_state":"computed","paper":{"title":"Teach Old SAEs New Domain Tricks with Boosting","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Daniil Gavrilov, Daniil Laptev, Gleb Gerasimov, Nikita Balagansky, Nikita Koriagin, Yaroslav Aksenov","submitted_at":"2025-07-17T10:57:49Z","abstract_excerpt":"Sparse Autoencoders have emerged as powerful tools for interpreting the internal representations of Large Language Models, yet they often fail to capture domain-specific features not prevalent in their training corpora. This paper introduces a residual learning approach that addresses this feature blindness without requiring complete retraining. We propose training a secondary SAE specifically to model the reconstruction error of a pretrained SAE on domain-specific texts, effectively capturing features missed by the primary model. By summing the outputs of both models during inference, we demo"},"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":"2507.12990","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-17T10:57:49Z","cross_cats_sorted":["cs.AI","cs.CL"],"title_canon_sha256":"1c1a282aba4996c7ec2f03438536d99af432b3b9a886b82356c2c1fbe743cefd","abstract_canon_sha256":"99527cb43ca485b4260de4669e28c7ce738370b7bdd261d4d44dc0e6618aebcf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:38:42.740623Z","signature_b64":"4U3cDrIwjLBdr9BwDRspfVM60BGJKE3WUf2omgAqvxyi8lv2NVliOA3/r1yjJ1d8Q30WPr5Wh4j1H1QJLIO8Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6053e668ff65c34840ccc7f492f595e57033e5f2292ff90555ebe31500c44c01","last_reissued_at":"2026-07-05T11:38:42.739995Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:38:42.739995Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Teach Old SAEs New Domain Tricks with Boosting","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL"],"primary_cat":"cs.LG","authors_text":"Daniil Gavrilov, Daniil Laptev, Gleb Gerasimov, Nikita Balagansky, Nikita Koriagin, Yaroslav Aksenov","submitted_at":"2025-07-17T10:57:49Z","abstract_excerpt":"Sparse Autoencoders have emerged as powerful tools for interpreting the internal representations of Large Language Models, yet they often fail to capture domain-specific features not prevalent in their training corpora. This paper introduces a residual learning approach that addresses this feature blindness without requiring complete retraining. We propose training a secondary SAE specifically to model the reconstruction error of a pretrained SAE on domain-specific texts, effectively capturing features missed by the primary model. By summing the outputs of both models during inference, we demo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.12990","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/2507.12990/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":"2507.12990","created_at":"2026-07-05T11:38:42.740125+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.12990v1","created_at":"2026-07-05T11:38:42.740125+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.12990","created_at":"2026-07-05T11:38:42.740125+00:00"},{"alias_kind":"pith_short_12","alias_value":"MBJ6M2H7MXBU","created_at":"2026-07-05T11:38:42.740125+00:00"},{"alias_kind":"pith_short_16","alias_value":"MBJ6M2H7MXBUQQGM","created_at":"2026-07-05T11:38:42.740125+00:00"},{"alias_kind":"pith_short_8","alias_value":"MBJ6M2H7","created_at":"2026-07-05T11:38:42.740125+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.21849","citing_title":"Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MBJ6M2H7MXBUQQGMY72JF5MV4V","json":"https://pith.science/pith/MBJ6M2H7MXBUQQGMY72JF5MV4V.json","graph_json":"https://pith.science/api/pith-number/MBJ6M2H7MXBUQQGMY72JF5MV4V/graph.json","events_json":"https://pith.science/api/pith-number/MBJ6M2H7MXBUQQGMY72JF5MV4V/events.json","paper":"https://pith.science/paper/MBJ6M2H7"},"agent_actions":{"view_html":"https://pith.science/pith/MBJ6M2H7MXBUQQGMY72JF5MV4V","download_json":"https://pith.science/pith/MBJ6M2H7MXBUQQGMY72JF5MV4V.json","view_paper":"https://pith.science/paper/MBJ6M2H7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.12990&json=true","fetch_graph":"https://pith.science/api/pith-number/MBJ6M2H7MXBUQQGMY72JF5MV4V/graph.json","fetch_events":"https://pith.science/api/pith-number/MBJ6M2H7MXBUQQGMY72JF5MV4V/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MBJ6M2H7MXBUQQGMY72JF5MV4V/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MBJ6M2H7MXBUQQGMY72JF5MV4V/action/storage_attestation","attest_author":"https://pith.science/pith/MBJ6M2H7MXBUQQGMY72JF5MV4V/action/author_attestation","sign_citation":"https://pith.science/pith/MBJ6M2H7MXBUQQGMY72JF5MV4V/action/citation_signature","submit_replication":"https://pith.science/pith/MBJ6M2H7MXBUQQGMY72JF5MV4V/action/replication_record"}},"created_at":"2026-07-05T11:38:42.740125+00:00","updated_at":"2026-07-05T11:38:42.740125+00:00"}