{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:ZMNEGPEZBD5O7FHQUA5QKDDWQS","short_pith_number":"pith:ZMNEGPEZ","canonical_record":{"source":{"id":"2406.17759","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-06-25T17:43:13Z","cross_cats_sorted":[],"title_canon_sha256":"606595001ffb10c1c2e4f814ea81025dfedc29fcbe1b4387d4c444715258125e","abstract_canon_sha256":"46905bf89efcc9e8eaefbb26d53f9acdeeb3a2e8c12b9d98d80fa56999a5e0e0"},"schema_version":"1.0"},"canonical_sha256":"cb1a433c9908faef94f0a03b050c7684b09e4e2ebe0279b596401cbaec0be18f","source":{"kind":"arxiv","id":"2406.17759","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2406.17759","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"arxiv_version","alias_value":"2406.17759v1","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.17759","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"pith_short_12","alias_value":"ZMNEGPEZBD5O","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"pith_short_16","alias_value":"ZMNEGPEZBD5O7FHQ","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"pith_short_8","alias_value":"ZMNEGPEZ","created_at":"2026-07-05T08:36:39Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:ZMNEGPEZBD5O7FHQUA5QKDDWQS","target":"record","payload":{"canonical_record":{"source":{"id":"2406.17759","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-06-25T17:43:13Z","cross_cats_sorted":[],"title_canon_sha256":"606595001ffb10c1c2e4f814ea81025dfedc29fcbe1b4387d4c444715258125e","abstract_canon_sha256":"46905bf89efcc9e8eaefbb26d53f9acdeeb3a2e8c12b9d98d80fa56999a5e0e0"},"schema_version":"1.0"},"canonical_sha256":"cb1a433c9908faef94f0a03b050c7684b09e4e2ebe0279b596401cbaec0be18f","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:36:39.535629Z","signature_b64":"/iClwkoML9mqgiSTeAXEPrcZqY5etQuycG++w7N3t2fvOs558kd3h3TJ3/pOTLDEAf4+c72+XX0UGaLKDQEfBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cb1a433c9908faef94f0a03b050c7684b09e4e2ebe0279b596401cbaec0be18f","last_reissued_at":"2026-07-05T08:36:39.535131Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:36:39.535131Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2406.17759","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T08:36:39Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"2CftPIRPJcpHZVA9oZB9taJAFpv8VqiP5idScAxKtjuFgOtp3sZXGsU2HQrDo+kaEd8TdmZBxJnVqngIK8zJAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-17T19:38:26.908835Z"},"content_sha256":"896b458222453be9da7f17197184f5ed0dba533ceddee8e5e2b8e55e34b42ec0","schema_version":"1.0","event_id":"sha256:896b458222453be9da7f17197184f5ed0dba533ceddee8e5e2b8e55e34b42ec0"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:ZMNEGPEZBD5O7FHQUA5QKDDWQS","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Interpreting Attention Layer Outputs with Sparse Autoencoders","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Arthur Conmy, Connor Kissane, Joseph Isaac Bloom, Neel Nanda, Robert Krzyzanowski","submitted_at":"2024-06-25T17:43:13Z","abstract_excerpt":"Decomposing model activations into interpretable components is a key open problem in mechanistic interpretability. Sparse autoencoders (SAEs) are a popular method for decomposing the internal activations of trained transformers into sparse, interpretable features, and have been applied to MLP layers and the residual stream. In this work we train SAEs on attention layer outputs and show that also here SAEs find a sparse, interpretable decomposition. We demonstrate this on transformers from several model families and up to 2B parameters.\n  We perform a qualitative study of the features computed "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.17759","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/2406.17759/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T08:36:39Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"AqPYqw5ZbXZMgoRh6G7NRxmOkAFDmpQm8QKBmh+ac9Sixx1HScH8rpweeeoZSOMmSRqtE0DFfz+3wrfO7ksmBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-17T19:38:26.909365Z"},"content_sha256":"e88b6bc441cea602a3db3d76ce3ad7dfbca70602cac0b587a61cfb529b1c1af5","schema_version":"1.0","event_id":"sha256:e88b6bc441cea602a3db3d76ce3ad7dfbca70602cac0b587a61cfb529b1c1af5"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/ZMNEGPEZBD5O7FHQUA5QKDDWQS/bundle.json","state_url":"https://pith.science/pith/ZMNEGPEZBD5O7FHQUA5QKDDWQS/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/ZMNEGPEZBD5O7FHQUA5QKDDWQS/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-17T19:38:26Z","links":{"resolver":"https://pith.science/pith/ZMNEGPEZBD5O7FHQUA5QKDDWQS","bundle":"https://pith.science/pith/ZMNEGPEZBD5O7FHQUA5QKDDWQS/bundle.json","state":"https://pith.science/pith/ZMNEGPEZBD5O7FHQUA5QKDDWQS/state.json","well_known_bundle":"https://pith.science/.well-known/pith/ZMNEGPEZBD5O7FHQUA5QKDDWQS/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:ZMNEGPEZBD5O7FHQUA5QKDDWQS","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"46905bf89efcc9e8eaefbb26d53f9acdeeb3a2e8c12b9d98d80fa56999a5e0e0","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-06-25T17:43:13Z","title_canon_sha256":"606595001ffb10c1c2e4f814ea81025dfedc29fcbe1b4387d4c444715258125e"},"schema_version":"1.0","source":{"id":"2406.17759","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2406.17759","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"arxiv_version","alias_value":"2406.17759v1","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.17759","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"pith_short_12","alias_value":"ZMNEGPEZBD5O","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"pith_short_16","alias_value":"ZMNEGPEZBD5O7FHQ","created_at":"2026-07-05T08:36:39Z"},{"alias_kind":"pith_short_8","alias_value":"ZMNEGPEZ","created_at":"2026-07-05T08:36:39Z"}],"graph_snapshots":[{"event_id":"sha256:e88b6bc441cea602a3db3d76ce3ad7dfbca70602cac0b587a61cfb529b1c1af5","target":"graph","created_at":"2026-07-05T08:36:39Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2406.17759/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Decomposing model activations into interpretable components is a key open problem in mechanistic interpretability. Sparse autoencoders (SAEs) are a popular method for decomposing the internal activations of trained transformers into sparse, interpretable features, and have been applied to MLP layers and the residual stream. In this work we train SAEs on attention layer outputs and show that also here SAEs find a sparse, interpretable decomposition. We demonstrate this on transformers from several model families and up to 2B parameters.\n  We perform a qualitative study of the features computed ","authors_text":"Arthur Conmy, Connor Kissane, Joseph Isaac Bloom, Neel Nanda, Robert Krzyzanowski","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-06-25T17:43:13Z","title":"Interpreting Attention Layer Outputs with Sparse Autoencoders"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.17759","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:896b458222453be9da7f17197184f5ed0dba533ceddee8e5e2b8e55e34b42ec0","target":"record","created_at":"2026-07-05T08:36:39Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"46905bf89efcc9e8eaefbb26d53f9acdeeb3a2e8c12b9d98d80fa56999a5e0e0","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-06-25T17:43:13Z","title_canon_sha256":"606595001ffb10c1c2e4f814ea81025dfedc29fcbe1b4387d4c444715258125e"},"schema_version":"1.0","source":{"id":"2406.17759","kind":"arxiv","version":1}},"canonical_sha256":"cb1a433c9908faef94f0a03b050c7684b09e4e2ebe0279b596401cbaec0be18f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"cb1a433c9908faef94f0a03b050c7684b09e4e2ebe0279b596401cbaec0be18f","first_computed_at":"2026-07-05T08:36:39.535131Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T08:36:39.535131Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"/iClwkoML9mqgiSTeAXEPrcZqY5etQuycG++w7N3t2fvOs558kd3h3TJ3/pOTLDEAf4+c72+XX0UGaLKDQEfBQ==","signature_status":"signed_v1","signed_at":"2026-07-05T08:36:39.535629Z","signed_message":"canonical_sha256_bytes"},"source_id":"2406.17759","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:896b458222453be9da7f17197184f5ed0dba533ceddee8e5e2b8e55e34b42ec0","sha256:e88b6bc441cea602a3db3d76ce3ad7dfbca70602cac0b587a61cfb529b1c1af5"],"state_sha256":"bbb6fb2a85e5eb331015b2ceb1ce31d36db24fbe2d43d01f6a47f16416d060a0"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Kmi38BYvRxXBS4PDXXKlyH2qhN9IsdRSsfW0RObW8k6m9cwwiS5fvMTxF6y5gFQfXwr9MFpHMcJY0ln0/jMXAQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-17T19:38:26.915391Z","bundle_sha256":"956b2c87f65859c12339be97502528f6263edfbae0d598c7fda1dfe5816120f0"}}