{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:NKIOL3VWQLF5GJERKQPTB6XR55","short_pith_number":"pith:NKIOL3VW","schema_version":"1.0","canonical_sha256":"6a90e5eeb682cbd32491541f30faf1ef477d3d3005c31d1aed28163097b64b3f","source":{"kind":"arxiv","id":"2405.08366","version":3},"attestation_state":"computed","paper":{"title":"Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aleksandar Makelov, George Lange, Neel Nanda","submitted_at":"2024-05-14T07:07:13Z","abstract_excerpt":"Disentangling model activations into meaningful features is a central problem in interpretability. However, the absence of ground-truth for these features in realistic scenarios makes validating recent approaches, such as sparse dictionary learning, elusive. To address this challenge, we propose a framework for evaluating feature dictionaries in the context of specific tasks, by comparing them against \\emph{supervised} feature dictionaries. First, we demonstrate that supervised dictionaries achieve excellent approximation, control, and interpretability of model computations on the task. Second"},"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":"2405.08366","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-05-14T07:07:13Z","cross_cats_sorted":[],"title_canon_sha256":"133a1ffbc93acca1c61eac356087fe88b2e80cde46f2d8db55cc16908b6d3b16","abstract_canon_sha256":"a74fa01587040fadcb768e1396108809297364cea3761d149500ebfd39e5952d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:20:54.646228Z","signature_b64":"Bbf7lj3K+FqH8kdY7B5Mc4gx/alNyG0QR8T1t8FZA0ZzWxF79KqLUc+1tYNfk+r4eXiOEyP+lO8HVEOXYJW0CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6a90e5eeb682cbd32491541f30faf1ef477d3d3005c31d1aed28163097b64b3f","last_reissued_at":"2026-07-05T08:20:54.645739Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:20:54.645739Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Aleksandar Makelov, George Lange, Neel Nanda","submitted_at":"2024-05-14T07:07:13Z","abstract_excerpt":"Disentangling model activations into meaningful features is a central problem in interpretability. However, the absence of ground-truth for these features in realistic scenarios makes validating recent approaches, such as sparse dictionary learning, elusive. To address this challenge, we propose a framework for evaluating feature dictionaries in the context of specific tasks, by comparing them against \\emph{supervised} feature dictionaries. First, we demonstrate that supervised dictionaries achieve excellent approximation, control, and interpretability of model computations on the task. Second"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.08366","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/2405.08366/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":"2405.08366","created_at":"2026-07-05T08:20:54.645799+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.08366v3","created_at":"2026-07-05T08:20:54.645799+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.08366","created_at":"2026-07-05T08:20:54.645799+00:00"},{"alias_kind":"pith_short_12","alias_value":"NKIOL3VWQLF5","created_at":"2026-07-05T08:20:54.645799+00:00"},{"alias_kind":"pith_short_16","alias_value":"NKIOL3VWQLF5GJER","created_at":"2026-07-05T08:20:54.645799+00:00"},{"alias_kind":"pith_short_8","alias_value":"NKIOL3VW","created_at":"2026-07-05T08:20:54.645799+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":9,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07316","citing_title":"Mechanistic Interpretability for Neural Networks: Circuits, Sparse Features and Symbolic Reasoning","ref_index":65,"is_internal_anchor":true},{"citing_arxiv_id":"2605.28149","citing_title":"Sign-Aware Gated Sparse Autoencoders: Modeling Anticorrelated Features with Bi-Jump-ReLU Activations","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29889","citing_title":"Internal Representation, Not Clinical Knowledge: Where Apparent LLM Triage Failures Originate","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18789","citing_title":"Features have life history. And we should care","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18229","citing_title":"Are Sparse Autoencoder Benchmarks Reliable?","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2512.19115","citing_title":"Generative Giants, Retrieval Weaklings: Why do Multimodal Large Language Models Fail at Multimodal Retrieval?","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12874","citing_title":"Descriptive Collision in Sparse Autoencoder Auto-Interpretability: When One Explanation Describes Many Features","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2406.11717","citing_title":"Refusal in Language Models Is Mediated by a Single Direction","ref_index":154,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19260","citing_title":"Understanding the Mechanism of Altruism in Large Language Models","ref_index":244,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NKIOL3VWQLF5GJERKQPTB6XR55","json":"https://pith.science/pith/NKIOL3VWQLF5GJERKQPTB6XR55.json","graph_json":"https://pith.science/api/pith-number/NKIOL3VWQLF5GJERKQPTB6XR55/graph.json","events_json":"https://pith.science/api/pith-number/NKIOL3VWQLF5GJERKQPTB6XR55/events.json","paper":"https://pith.science/paper/NKIOL3VW"},"agent_actions":{"view_html":"https://pith.science/pith/NKIOL3VWQLF5GJERKQPTB6XR55","download_json":"https://pith.science/pith/NKIOL3VWQLF5GJERKQPTB6XR55.json","view_paper":"https://pith.science/paper/NKIOL3VW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.08366&json=true","fetch_graph":"https://pith.science/api/pith-number/NKIOL3VWQLF5GJERKQPTB6XR55/graph.json","fetch_events":"https://pith.science/api/pith-number/NKIOL3VWQLF5GJERKQPTB6XR55/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NKIOL3VWQLF5GJERKQPTB6XR55/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NKIOL3VWQLF5GJERKQPTB6XR55/action/storage_attestation","attest_author":"https://pith.science/pith/NKIOL3VWQLF5GJERKQPTB6XR55/action/author_attestation","sign_citation":"https://pith.science/pith/NKIOL3VWQLF5GJERKQPTB6XR55/action/citation_signature","submit_replication":"https://pith.science/pith/NKIOL3VWQLF5GJERKQPTB6XR55/action/replication_record"}},"created_at":"2026-07-05T08:20:54.645799+00:00","updated_at":"2026-07-05T08:20:54.645799+00:00"}