{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:NPHMDM3JNPAVX4NMEFIBLAYYJW","short_pith_number":"pith:NPHMDM3J","canonical_record":{"source":{"id":"2504.08729","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-04-11T17:56:09Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"3540603e5628ac1e1d135780fa5b709a2eb135263cebca0c8164569036c9b11f","abstract_canon_sha256":"2f6d0fd2b8c7c53cd242f45c2f51b6e537bc0aa1b0c0e143411093fdf5f9c369"},"schema_version":"1.0"},"canonical_sha256":"6bcec1b3696bc15bf1ac21501583184dad00ec776a08b30d629b90c4ef5e4eb5","source":{"kind":"arxiv","id":"2504.08729","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.08729","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"arxiv_version","alias_value":"2504.08729v1","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.08729","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"pith_short_12","alias_value":"NPHMDM3JNPAV","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"pith_short_16","alias_value":"NPHMDM3JNPAVX4NM","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"pith_short_8","alias_value":"NPHMDM3J","created_at":"2026-07-05T10:47:53Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:NPHMDM3JNPAVX4NMEFIBLAYYJW","target":"record","payload":{"canonical_record":{"source":{"id":"2504.08729","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-04-11T17:56:09Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"3540603e5628ac1e1d135780fa5b709a2eb135263cebca0c8164569036c9b11f","abstract_canon_sha256":"2f6d0fd2b8c7c53cd242f45c2f51b6e537bc0aa1b0c0e143411093fdf5f9c369"},"schema_version":"1.0"},"canonical_sha256":"6bcec1b3696bc15bf1ac21501583184dad00ec776a08b30d629b90c4ef5e4eb5","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:47:53.284726Z","signature_b64":"K8NdY4ANTbjogbLTioVTSI3U1iIHwMbRtfivk/CRhhGOm0/xG53Y4kkwIz4iDKfZFEoWkOc6VMc1fYKix/pJCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6bcec1b3696bc15bf1ac21501583184dad00ec776a08b30d629b90c4ef5e4eb5","last_reissued_at":"2026-07-05T10:47:53.284242Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:47:53.284242Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2504.08729","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":"2026-07-05T10:47:53Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Jp+lQtvTmJLn8N+5V0uFovAffVbp5Sdmp0HaA1C88UNcK1Qei3v+ZfrGVEF178sDgrKs3T3OEkTZQ0d0QKbsCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-05T01:27:45.333925Z"},"content_sha256":"4c312287f74a60c2d68806b44efe3b98ca70c47f18deaad62dc716528cd55a3f","schema_version":"1.0","event_id":"sha256:4c312287f74a60c2d68806b44efe3b98ca70c47f18deaad62dc716528cd55a3f"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:NPHMDM3JNPAVX4NMEFIBLAYYJW","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Steering CLIP's vision transformer with sparse autoencoders","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Blake Aaron Richards, Danilo Bzdok, Ethan Goldfarb, Lorenz Hufe, Praneet Suresh, Robert Graham, Sonia Joseph, Wojciech Samek, Yossi Gandelsman","submitted_at":"2025-04-11T17:56:09Z","abstract_excerpt":"While vision models are highly capable, their internal mechanisms remain poorly understood -- a challenge which sparse autoencoders (SAEs) have helped address in language, but which remains underexplored in vision. We address this gap by training SAEs on CLIP's vision transformer and uncover key differences between vision and language processing, including distinct sparsity patterns for SAEs trained across layers and token types. We then provide the first systematic analysis on the steerability of CLIP's vision transformer by introducing metrics to quantify how precisely SAE features can be st"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.08729","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/2504.08729/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":"2026-07-05T10:47:53Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Mgc6cTpdpN60kLbzdZUuHdg3fNJC9jkcAKf1hu4udLXqQbraRsuMHKgqmlBh6Y/NcyXDmMHy2SSKwciw8xDVBg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-05T01:27:45.335103Z"},"content_sha256":"ef57a3ee59294c79c40f54fe424af97ddf82a745bfa19c22b9761a8aaec60755","schema_version":"1.0","event_id":"sha256:ef57a3ee59294c79c40f54fe424af97ddf82a745bfa19c22b9761a8aaec60755"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/NPHMDM3JNPAVX4NMEFIBLAYYJW/bundle.json","state_url":"https://pith.science/pith/NPHMDM3JNPAVX4NMEFIBLAYYJW/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/NPHMDM3JNPAVX4NMEFIBLAYYJW/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":"2026-08-05T01:27:45Z","links":{"resolver":"https://pith.science/pith/NPHMDM3JNPAVX4NMEFIBLAYYJW","bundle":"https://pith.science/pith/NPHMDM3JNPAVX4NMEFIBLAYYJW/bundle.json","state":"https://pith.science/pith/NPHMDM3JNPAVX4NMEFIBLAYYJW/state.json","well_known_bundle":"https://pith.science/.well-known/pith/NPHMDM3JNPAVX4NMEFIBLAYYJW/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:NPHMDM3JNPAVX4NMEFIBLAYYJW","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":"2f6d0fd2b8c7c53cd242f45c2f51b6e537bc0aa1b0c0e143411093fdf5f9c369","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-04-11T17:56:09Z","title_canon_sha256":"3540603e5628ac1e1d135780fa5b709a2eb135263cebca0c8164569036c9b11f"},"schema_version":"1.0","source":{"id":"2504.08729","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.08729","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"arxiv_version","alias_value":"2504.08729v1","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.08729","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"pith_short_12","alias_value":"NPHMDM3JNPAV","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"pith_short_16","alias_value":"NPHMDM3JNPAVX4NM","created_at":"2026-07-05T10:47:53Z"},{"alias_kind":"pith_short_8","alias_value":"NPHMDM3J","created_at":"2026-07-05T10:47:53Z"}],"graph_snapshots":[{"event_id":"sha256:ef57a3ee59294c79c40f54fe424af97ddf82a745bfa19c22b9761a8aaec60755","target":"graph","created_at":"2026-07-05T10:47:53Z","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/2504.08729/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"While vision models are highly capable, their internal mechanisms remain poorly understood -- a challenge which sparse autoencoders (SAEs) have helped address in language, but which remains underexplored in vision. We address this gap by training SAEs on CLIP's vision transformer and uncover key differences between vision and language processing, including distinct sparsity patterns for SAEs trained across layers and token types. We then provide the first systematic analysis on the steerability of CLIP's vision transformer by introducing metrics to quantify how precisely SAE features can be st","authors_text":"Blake Aaron Richards, Danilo Bzdok, Ethan Goldfarb, Lorenz Hufe, Praneet Suresh, Robert Graham, Sonia Joseph, Wojciech Samek, Yossi Gandelsman","cross_cats":["cs.AI","cs.LG"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-04-11T17:56:09Z","title":"Steering CLIP's vision transformer with sparse autoencoders"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.08729","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":"sha256:4c312287f74a60c2d68806b44efe3b98ca70c47f18deaad62dc716528cd55a3f","target":"record","created_at":"2026-07-05T10:47:53Z","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":"2f6d0fd2b8c7c53cd242f45c2f51b6e537bc0aa1b0c0e143411093fdf5f9c369","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-04-11T17:56:09Z","title_canon_sha256":"3540603e5628ac1e1d135780fa5b709a2eb135263cebca0c8164569036c9b11f"},"schema_version":"1.0","source":{"id":"2504.08729","kind":"arxiv","version":1}},"canonical_sha256":"6bcec1b3696bc15bf1ac21501583184dad00ec776a08b30d629b90c4ef5e4eb5","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"6bcec1b3696bc15bf1ac21501583184dad00ec776a08b30d629b90c4ef5e4eb5","first_computed_at":"2026-07-05T10:47:53.284242Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T10:47:53.284242Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"K8NdY4ANTbjogbLTioVTSI3U1iIHwMbRtfivk/CRhhGOm0/xG53Y4kkwIz4iDKfZFEoWkOc6VMc1fYKix/pJCA==","signature_status":"signed_v1","signed_at":"2026-07-05T10:47:53.284726Z","signed_message":"canonical_sha256_bytes"},"source_id":"2504.08729","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:4c312287f74a60c2d68806b44efe3b98ca70c47f18deaad62dc716528cd55a3f","sha256:ef57a3ee59294c79c40f54fe424af97ddf82a745bfa19c22b9761a8aaec60755"],"state_sha256":"81b6cc8b9437d273b0bd72627fa155fa3ee4337dec5658593929d94a41c9343c"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"iyOJ9FeIodDYmJscTYsiK7oTEYa+NE7zFISbh1UttDD5aBTibHyXEwWjFUoeUlRNXmgZxrVrpxtEQVFmiAP4AA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-05T01:27:45.342723Z","bundle_sha256":"191765daf1b509467eaddb0717fa218b363af6c5b35207379b353eff5bc40f68"}}