{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:P3LPV2VN3E3CQRIUPWPUPP7MWC","short_pith_number":"pith:P3LPV2VN","canonical_record":{"source":{"id":"2505.14071","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-20T08:23:08Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"279e7ebef37683573c16be845eae704e4ead4e621b6f22e718cd9441978dd40e","abstract_canon_sha256":"ee211ba8735843d7c84eb4e4dadfcf4ef4ca0281655c40eb586654e12ad80abc"},"schema_version":"1.0"},"canonical_sha256":"7ed6faeaadd9362845147d9f47bfecb098a08034b61858f5710ee7f10046f92e","source":{"kind":"arxiv","id":"2505.14071","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.14071","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"arxiv_version","alias_value":"2505.14071v1","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.14071","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"pith_short_12","alias_value":"P3LPV2VN3E3C","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"pith_short_16","alias_value":"P3LPV2VN3E3CQRIU","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"pith_short_8","alias_value":"P3LPV2VN","created_at":"2026-07-05T11:05:55Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:P3LPV2VN3E3CQRIUPWPUPP7MWC","target":"record","payload":{"canonical_record":{"source":{"id":"2505.14071","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-20T08:23:08Z","cross_cats_sorted":["cs.CL","cs.CV"],"title_canon_sha256":"279e7ebef37683573c16be845eae704e4ead4e621b6f22e718cd9441978dd40e","abstract_canon_sha256":"ee211ba8735843d7c84eb4e4dadfcf4ef4ca0281655c40eb586654e12ad80abc"},"schema_version":"1.0"},"canonical_sha256":"7ed6faeaadd9362845147d9f47bfecb098a08034b61858f5710ee7f10046f92e","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:05:55.690003Z","signature_b64":"J/Ehm2xl/VeDv1dr1WWiA/D4qYYDAI3KwFaOaJQ/hZxTjU5/bXHdsPM42i2Eq4aEbcormNQnnbA+wEc13V7ZDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7ed6faeaadd9362845147d9f47bfecb098a08034b61858f5710ee7f10046f92e","last_reissued_at":"2026-07-05T11:05:55.689515Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:05:55.689515Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2505.14071","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":"2026-07-05T11:05:55Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"w04HcnVE57t3cTbb4Z2zxs8VHpnzY2/1S90AmrJGK0JWah1TwMJPtMso62s29wGOv8tWlhSZT6VHyoh3UQCMBA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-05T13:13:27.114547Z"},"content_sha256":"864817e8d811c7255a6bbb3c0131cbcaa3db99b00062b028c72319567b909381","schema_version":"1.0","event_id":"sha256:864817e8d811c7255a6bbb3c0131cbcaa3db99b00062b028c72319567b909381"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:P3LPV2VN3E3CQRIUPWPUPP7MWC","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Textual Steering Vectors Can Improve Visual Understanding in Multimodal Large Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Dani Yogatama, Deqing Fu, Julian Asilis, Ollie Liu, Robin Jia, Vatsal Sharan, Willie Neiswanger, Woody Haosheng Gan","submitted_at":"2025-05-20T08:23:08Z","abstract_excerpt":"Steering methods have emerged as effective and targeted tools for guiding large language models' (LLMs) behavior without modifying their parameters. Multimodal large language models (MLLMs), however, do not currently enjoy the same suite of techniques, due in part to their recency and architectural diversity. Inspired by this gap, we investigate whether MLLMs can be steered using vectors derived from their text-only LLM backbone, via sparse autoencoders (SAEs), mean shift, and linear probing. We find that text-derived steering consistently enhances multimodal accuracy across diverse MLLM archi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.14071","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/2505.14071/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":"2026-07-05T11:05:55Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"yoedEoCl+dqC4YIe4DSvF+67K8U+zGq7nwq64yAADs9YS+ALVUoxyWWrdd06JqJ41XFUbjZr8a6OAPYVOu2RCA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-05T13:13:27.115659Z"},"content_sha256":"6867d3c928d9f1abb1425d00ba6d38fcad13fac8dc7b9707f43e008c44d9a79e","schema_version":"1.0","event_id":"sha256:6867d3c928d9f1abb1425d00ba6d38fcad13fac8dc7b9707f43e008c44d9a79e"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/P3LPV2VN3E3CQRIUPWPUPP7MWC/bundle.json","state_url":"https://pith.science/pith/P3LPV2VN3E3CQRIUPWPUPP7MWC/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/P3LPV2VN3E3CQRIUPWPUPP7MWC/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":"2026-08-05T13:13:27Z","links":{"resolver":"https://pith.science/pith/P3LPV2VN3E3CQRIUPWPUPP7MWC","bundle":"https://pith.science/pith/P3LPV2VN3E3CQRIUPWPUPP7MWC/bundle.json","state":"https://pith.science/pith/P3LPV2VN3E3CQRIUPWPUPP7MWC/state.json","well_known_bundle":"https://pith.science/.well-known/pith/P3LPV2VN3E3CQRIUPWPUPP7MWC/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:P3LPV2VN3E3CQRIUPWPUPP7MWC","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":"ee211ba8735843d7c84eb4e4dadfcf4ef4ca0281655c40eb586654e12ad80abc","cross_cats_sorted":["cs.CL","cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-20T08:23:08Z","title_canon_sha256":"279e7ebef37683573c16be845eae704e4ead4e621b6f22e718cd9441978dd40e"},"schema_version":"1.0","source":{"id":"2505.14071","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.14071","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"arxiv_version","alias_value":"2505.14071v1","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.14071","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"pith_short_12","alias_value":"P3LPV2VN3E3C","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"pith_short_16","alias_value":"P3LPV2VN3E3CQRIU","created_at":"2026-07-05T11:05:55Z"},{"alias_kind":"pith_short_8","alias_value":"P3LPV2VN","created_at":"2026-07-05T11:05:55Z"}],"graph_snapshots":[{"event_id":"sha256:6867d3c928d9f1abb1425d00ba6d38fcad13fac8dc7b9707f43e008c44d9a79e","target":"graph","created_at":"2026-07-05T11:05:55Z","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/2505.14071/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Steering methods have emerged as effective and targeted tools for guiding large language models' (LLMs) behavior without modifying their parameters. Multimodal large language models (MLLMs), however, do not currently enjoy the same suite of techniques, due in part to their recency and architectural diversity. Inspired by this gap, we investigate whether MLLMs can be steered using vectors derived from their text-only LLM backbone, via sparse autoencoders (SAEs), mean shift, and linear probing. We find that text-derived steering consistently enhances multimodal accuracy across diverse MLLM archi","authors_text":"Dani Yogatama, Deqing Fu, Julian Asilis, Ollie Liu, Robin Jia, Vatsal Sharan, Willie Neiswanger, Woody Haosheng Gan","cross_cats":["cs.CL","cs.CV"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-20T08:23:08Z","title":"Textual Steering Vectors Can Improve Visual Understanding in Multimodal Large Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.14071","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":"sha256:864817e8d811c7255a6bbb3c0131cbcaa3db99b00062b028c72319567b909381","target":"record","created_at":"2026-07-05T11:05:55Z","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":"ee211ba8735843d7c84eb4e4dadfcf4ef4ca0281655c40eb586654e12ad80abc","cross_cats_sorted":["cs.CL","cs.CV"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-20T08:23:08Z","title_canon_sha256":"279e7ebef37683573c16be845eae704e4ead4e621b6f22e718cd9441978dd40e"},"schema_version":"1.0","source":{"id":"2505.14071","kind":"arxiv","version":1}},"canonical_sha256":"7ed6faeaadd9362845147d9f47bfecb098a08034b61858f5710ee7f10046f92e","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"7ed6faeaadd9362845147d9f47bfecb098a08034b61858f5710ee7f10046f92e","first_computed_at":"2026-07-05T11:05:55.689515Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:05:55.689515Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"J/Ehm2xl/VeDv1dr1WWiA/D4qYYDAI3KwFaOaJQ/hZxTjU5/bXHdsPM42i2Eq4aEbcormNQnnbA+wEc13V7ZDA==","signature_status":"signed_v1","signed_at":"2026-07-05T11:05:55.690003Z","signed_message":"canonical_sha256_bytes"},"source_id":"2505.14071","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:864817e8d811c7255a6bbb3c0131cbcaa3db99b00062b028c72319567b909381","sha256:6867d3c928d9f1abb1425d00ba6d38fcad13fac8dc7b9707f43e008c44d9a79e"],"state_sha256":"ca61020bbc718c65961f47c8cdf5e09c9cc15bbc3eeb99bc295e5366fd5ba0cf"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ZBi3aK4uflX1dzVmHg7+QkFs8p8vCuZk9LratsefAONK13RMuBLbP252QIw24rdUj3x+31DJYMt1hy3kzWmzDg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-05T13:13:27.128685Z","bundle_sha256":"447f76b5eed426476f08e30c77f37a705e563b5851b4e1b8e98bf4da70b4f609"}}