{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:NZJQEFHLUS7UBAMBDTTDT3ZQMI","short_pith_number":"pith:NZJQEFHL","canonical_record":{"source":{"id":"2504.11695","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-04-16T01:40:06Z","cross_cats_sorted":["cs.CL","cs.MM"],"title_canon_sha256":"8a04e2c5143a3e41937a9581938fa157b0378caf548c2cd154ca230ac1b3cde0","abstract_canon_sha256":"08f9c2e5012230014364c016da86ab6b7ef12a64ae5b73f1588c945a8a8d668d"},"schema_version":"1.0"},"canonical_sha256":"6e530214eba4bf4081811ce639ef306229fe29ea73e56f12938e85518d58efdc","source":{"kind":"arxiv","id":"2504.11695","version":4},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.11695","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"arxiv_version","alias_value":"2504.11695v4","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.11695","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"pith_short_12","alias_value":"NZJQEFHLUS7U","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"pith_short_16","alias_value":"NZJQEFHLUS7UBAMB","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"pith_short_8","alias_value":"NZJQEFHL","created_at":"2026-07-05T11:57:35Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:NZJQEFHLUS7UBAMBDTTDT3ZQMI","target":"record","payload":{"canonical_record":{"source":{"id":"2504.11695","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-04-16T01:40:06Z","cross_cats_sorted":["cs.CL","cs.MM"],"title_canon_sha256":"8a04e2c5143a3e41937a9581938fa157b0378caf548c2cd154ca230ac1b3cde0","abstract_canon_sha256":"08f9c2e5012230014364c016da86ab6b7ef12a64ae5b73f1588c945a8a8d668d"},"schema_version":"1.0"},"canonical_sha256":"6e530214eba4bf4081811ce639ef306229fe29ea73e56f12938e85518d58efdc","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:57:35.920985Z","signature_b64":"/fzWufKf+xMiAcanGhIutG5azuXp82gf41a9DOzijyB3lBEPuf+LslYLNYc3xFZRe/DBpKlPzo9+KNo860ZMCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6e530214eba4bf4081811ce639ef306229fe29ea73e56f12938e85518d58efdc","last_reissued_at":"2026-07-05T11:57:35.920526Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:57:35.920526Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2504.11695","source_version":4,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:57:35Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"3dCJu95RGs7NJO97ZRnwnxzQ9nbqEDcIgzucsdeS+xN5pX2gGejQ55YErIB8QKQyrLZhR+0leuYKb2zanrcxDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-01T02:55:00.198844Z"},"content_sha256":"31f8f943fc004376b2c388a0f33318ab3bbc0b3c23ca9599e4713935973a02bc","schema_version":"1.0","event_id":"sha256:31f8f943fc004376b2c388a0f33318ab3bbc0b3c23ca9599e4713935973a02bc"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:NZJQEFHLUS7UBAMBDTTDT3ZQMI","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Interpreting the linear structure of vision-language model embedding spaces","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.MM"],"primary_cat":"cs.CV","authors_text":"Huangyuan Su, Isabel Papadimitriou, Sham Kakade, Stephanie Gil, Thomas Fel","submitted_at":"2025-04-16T01:40:06Z","abstract_excerpt":"Vision-language models encode images and text in a joint space, minimizing the distance between corresponding image and text pairs. How are language and images organized in this joint space, and how do the models encode meaning and modality? To investigate this, we train and release sparse autoencoders (SAEs) on the embedding spaces of four vision-language models (CLIP, SigLIP, SigLIP2, and AIMv2). SAEs approximate model embeddings as sparse linear combinations of learned directions, or \"concepts\". We find that, compared to other methods of linear feature learning, SAEs are better at reconstru"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.11695","kind":"arxiv","version":4},"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/2504.11695/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":"2026-07-05T11:57:35Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"/eDKtlYC5U0cJXgSOuxclX5GN0nY4KBU80s1mZRjTNCcktMgqwbkDllG5776OrSZum85dm6e3EzfU/AzNGJiCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-01T02:55:00.200117Z"},"content_sha256":"c33c9abbbd507eb588c8f72c6f18debcfd491b8ed8983c909d2c740b53ea35a4","schema_version":"1.0","event_id":"sha256:c33c9abbbd507eb588c8f72c6f18debcfd491b8ed8983c909d2c740b53ea35a4"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/NZJQEFHLUS7UBAMBDTTDT3ZQMI/bundle.json","state_url":"https://pith.science/pith/NZJQEFHLUS7UBAMBDTTDT3ZQMI/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/NZJQEFHLUS7UBAMBDTTDT3ZQMI/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":"2026-08-01T02:55:00Z","links":{"resolver":"https://pith.science/pith/NZJQEFHLUS7UBAMBDTTDT3ZQMI","bundle":"https://pith.science/pith/NZJQEFHLUS7UBAMBDTTDT3ZQMI/bundle.json","state":"https://pith.science/pith/NZJQEFHLUS7UBAMBDTTDT3ZQMI/state.json","well_known_bundle":"https://pith.science/.well-known/pith/NZJQEFHLUS7UBAMBDTTDT3ZQMI/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:NZJQEFHLUS7UBAMBDTTDT3ZQMI","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":"08f9c2e5012230014364c016da86ab6b7ef12a64ae5b73f1588c945a8a8d668d","cross_cats_sorted":["cs.CL","cs.MM"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-04-16T01:40:06Z","title_canon_sha256":"8a04e2c5143a3e41937a9581938fa157b0378caf548c2cd154ca230ac1b3cde0"},"schema_version":"1.0","source":{"id":"2504.11695","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2504.11695","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"arxiv_version","alias_value":"2504.11695v4","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.11695","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"pith_short_12","alias_value":"NZJQEFHLUS7U","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"pith_short_16","alias_value":"NZJQEFHLUS7UBAMB","created_at":"2026-07-05T11:57:35Z"},{"alias_kind":"pith_short_8","alias_value":"NZJQEFHL","created_at":"2026-07-05T11:57:35Z"}],"graph_snapshots":[{"event_id":"sha256:c33c9abbbd507eb588c8f72c6f18debcfd491b8ed8983c909d2c740b53ea35a4","target":"graph","created_at":"2026-07-05T11:57:35Z","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/2504.11695/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Vision-language models encode images and text in a joint space, minimizing the distance between corresponding image and text pairs. How are language and images organized in this joint space, and how do the models encode meaning and modality? To investigate this, we train and release sparse autoencoders (SAEs) on the embedding spaces of four vision-language models (CLIP, SigLIP, SigLIP2, and AIMv2). SAEs approximate model embeddings as sparse linear combinations of learned directions, or \"concepts\". We find that, compared to other methods of linear feature learning, SAEs are better at reconstru","authors_text":"Huangyuan Su, Isabel Papadimitriou, Sham Kakade, Stephanie Gil, Thomas Fel","cross_cats":["cs.CL","cs.MM"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-04-16T01:40:06Z","title":"Interpreting the linear structure of vision-language model embedding spaces"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.11695","kind":"arxiv","version":4},"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":"sha256:31f8f943fc004376b2c388a0f33318ab3bbc0b3c23ca9599e4713935973a02bc","target":"record","created_at":"2026-07-05T11:57:35Z","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":"08f9c2e5012230014364c016da86ab6b7ef12a64ae5b73f1588c945a8a8d668d","cross_cats_sorted":["cs.CL","cs.MM"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-04-16T01:40:06Z","title_canon_sha256":"8a04e2c5143a3e41937a9581938fa157b0378caf548c2cd154ca230ac1b3cde0"},"schema_version":"1.0","source":{"id":"2504.11695","kind":"arxiv","version":4}},"canonical_sha256":"6e530214eba4bf4081811ce639ef306229fe29ea73e56f12938e85518d58efdc","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"6e530214eba4bf4081811ce639ef306229fe29ea73e56f12938e85518d58efdc","first_computed_at":"2026-07-05T11:57:35.920526Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:57:35.920526Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"/fzWufKf+xMiAcanGhIutG5azuXp82gf41a9DOzijyB3lBEPuf+LslYLNYc3xFZRe/DBpKlPzo9+KNo860ZMCw==","signature_status":"signed_v1","signed_at":"2026-07-05T11:57:35.920985Z","signed_message":"canonical_sha256_bytes"},"source_id":"2504.11695","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:31f8f943fc004376b2c388a0f33318ab3bbc0b3c23ca9599e4713935973a02bc","sha256:c33c9abbbd507eb588c8f72c6f18debcfd491b8ed8983c909d2c740b53ea35a4"],"state_sha256":"c348703c59879209e0d12f5699a95e0bc75dc91e4867d8ff1e9e13cc033a1c53"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Uq/eDf83KiXld5EAxUklX5O5PhY3d49huPJ8r2LESH64MhSW140TM67xGGRz9LsNvgg0yu6rKpXjkth6VwuMCA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-01T02:55:00.208907Z","bundle_sha256":"cc678ff1a1d176edc60e81d0f54ceefe253f599367067e1a1e8d96ee9ee9ffcc"}}