{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:MJIY3TJCGQ5HQBXJ2SZ5RWSTLY","short_pith_number":"pith:MJIY3TJC","schema_version":"1.0","canonical_sha256":"62518dcd22343a7806e9d4b3d8da535e27fbfe66dc77e940fa20d48131952ceb","source":{"kind":"arxiv","id":"2402.12201","version":1},"attestation_state":"computed","paper":{"title":"Dictionary Learning Improves Patch-Free Circuit Discovery in Mechanistic Interpretability: A Case Study on Othello-GPT","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Qinyuan Cheng, Qiong Tang, Tianxiang Sun, Xipeng Qiu, Xuyang Ge, Zhengfu He","submitted_at":"2024-02-19T15:04:53Z","abstract_excerpt":"Sparse dictionary learning has been a rapidly growing technique in mechanistic interpretability to attack superposition and extract more human-understandable features from model activations. We ask a further question based on the extracted more monosemantic features: How do we recognize circuits connecting the enormous amount of dictionary features? We propose a circuit discovery framework alternative to activation patching. Our framework suffers less from out-of-distribution and proves to be more efficient in terms of asymptotic complexity. The basic unit in our framework is dictionary featur"},"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":"2402.12201","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-19T15:04:53Z","cross_cats_sorted":[],"title_canon_sha256":"62879ebcf45b8623736a6bc82ffd297c33652dd5e1b6d342839dd311dc9dbbda","abstract_canon_sha256":"481c2f805b2b6bb955abad00127e3bf4d268b247126625c6d63c8e6f20734967"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:46:50.232156Z","signature_b64":"jlXh1dIVQ1kVDR/L3vFQjokSU0GnHOzXtbe0MMkAiE4AXQdbGJTjmn8k2FWvZwFpu3trmJXWDmmriYSUKZbSBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"62518dcd22343a7806e9d4b3d8da535e27fbfe66dc77e940fa20d48131952ceb","last_reissued_at":"2026-07-05T07:46:50.231811Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:46:50.231811Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dictionary Learning Improves Patch-Free Circuit Discovery in Mechanistic Interpretability: A Case Study on Othello-GPT","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Qinyuan Cheng, Qiong Tang, Tianxiang Sun, Xipeng Qiu, Xuyang Ge, Zhengfu He","submitted_at":"2024-02-19T15:04:53Z","abstract_excerpt":"Sparse dictionary learning has been a rapidly growing technique in mechanistic interpretability to attack superposition and extract more human-understandable features from model activations. We ask a further question based on the extracted more monosemantic features: How do we recognize circuits connecting the enormous amount of dictionary features? We propose a circuit discovery framework alternative to activation patching. Our framework suffers less from out-of-distribution and proves to be more efficient in terms of asymptotic complexity. The basic unit in our framework is dictionary featur"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.12201","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/2402.12201/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":"2402.12201","created_at":"2026-07-05T07:46:50.231867+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.12201v1","created_at":"2026-07-05T07:46:50.231867+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.12201","created_at":"2026-07-05T07:46:50.231867+00:00"},{"alias_kind":"pith_short_12","alias_value":"MJIY3TJCGQ5H","created_at":"2026-07-05T07:46:50.231867+00:00"},{"alias_kind":"pith_short_16","alias_value":"MJIY3TJCGQ5HQBXJ","created_at":"2026-07-05T07:46:50.231867+00:00"},{"alias_kind":"pith_short_8","alias_value":"MJIY3TJC","created_at":"2026-07-05T07:46:50.231867+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.29358","citing_title":"Scaling Monosemanticity: Extracting Interpretable Features from Claude 3 Sonnet","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2601.14004","citing_title":"Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models","ref_index":106,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY","json":"https://pith.science/pith/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY.json","graph_json":"https://pith.science/api/pith-number/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY/graph.json","events_json":"https://pith.science/api/pith-number/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY/events.json","paper":"https://pith.science/paper/MJIY3TJC"},"agent_actions":{"view_html":"https://pith.science/pith/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY","download_json":"https://pith.science/pith/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY.json","view_paper":"https://pith.science/paper/MJIY3TJC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.12201&json=true","fetch_graph":"https://pith.science/api/pith-number/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY/graph.json","fetch_events":"https://pith.science/api/pith-number/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY/action/storage_attestation","attest_author":"https://pith.science/pith/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY/action/author_attestation","sign_citation":"https://pith.science/pith/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY/action/citation_signature","submit_replication":"https://pith.science/pith/MJIY3TJCGQ5HQBXJ2SZ5RWSTLY/action/replication_record"}},"created_at":"2026-07-05T07:46:50.231867+00:00","updated_at":"2026-07-05T07:46:50.231867+00:00"}