{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:OHACGJCNDB4YAPGULQ4US4BBVV","short_pith_number":"pith:OHACGJCN","schema_version":"1.0","canonical_sha256":"71c023244d1879803cd45c39497021ad6aa0cc5d28e86be17667e1ce68102afd","source":{"kind":"arxiv","id":"2410.02762","version":2},"attestation_state":"computed","paper":{"title":"Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Anish Kachinthaya, Nick Jiang, Suzie Petryk, Yossi Gandelsman","submitted_at":"2024-10-03T17:59:57Z","abstract_excerpt":"We investigate the internal representations of vision-language models (VLMs) to address hallucinations, a persistent challenge despite advances in model size and training. We project VLMs' internal image representations to their language vocabulary and observe more confident output probabilities on real objects than hallucinated objects. We additionally use these output probabilities to spatially localize real objects. Building on this approach, we introduce a knowledge erasure algorithm that removes hallucinations by linearly orthogonalizing image features with respect to hallucinated object "},"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":"2410.02762","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2024-10-03T17:59:57Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"2e1523b15f37e29a6c2706aae441e017cee7e01e5400205189963fa29d59705e","abstract_canon_sha256":"7a404df2905428f14b0773b71e2cce87b44985d5fe07ede0446e5a9a3528356a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:12:20.857253Z","signature_b64":"U7B3dXqt0rk6tW0x0th67d8ilJgROi+fQYh4V5OTec+e6flfXNl/lPt8PLXtki3OFMyZ8Cpl66ZYDl0QGBr9Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"71c023244d1879803cd45c39497021ad6aa0cc5d28e86be17667e1ce68102afd","last_reissued_at":"2026-07-05T10:12:20.856773Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:12:20.856773Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Anish Kachinthaya, Nick Jiang, Suzie Petryk, Yossi Gandelsman","submitted_at":"2024-10-03T17:59:57Z","abstract_excerpt":"We investigate the internal representations of vision-language models (VLMs) to address hallucinations, a persistent challenge despite advances in model size and training. We project VLMs' internal image representations to their language vocabulary and observe more confident output probabilities on real objects than hallucinated objects. We additionally use these output probabilities to spatially localize real objects. Building on this approach, we introduce a knowledge erasure algorithm that removes hallucinations by linearly orthogonalizing image features with respect to hallucinated object "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.02762","kind":"arxiv","version":2},"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/2410.02762/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":"2410.02762","created_at":"2026-07-05T10:12:20.856828+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.02762v2","created_at":"2026-07-05T10:12:20.856828+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.02762","created_at":"2026-07-05T10:12:20.856828+00:00"},{"alias_kind":"pith_short_12","alias_value":"OHACGJCNDB4Y","created_at":"2026-07-05T10:12:20.856828+00:00"},{"alias_kind":"pith_short_16","alias_value":"OHACGJCNDB4YAPGU","created_at":"2026-07-05T10:12:20.856828+00:00"},{"alias_kind":"pith_short_8","alias_value":"OHACGJCN","created_at":"2026-07-05T10:12:20.856828+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.06445","citing_title":"Analysis-by-Proxy: Localization Signals in VLMs Operating as Condition Encoders","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2606.20077","citing_title":"The Hidden Evolution of Disguised Visual Context inside the VLM","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2606.28273","citing_title":"Vision-Default, Prior-Override: Causal Mechanisms of Perception-Knowledge Conflict in Vision-Language Models","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2601.14004","citing_title":"Locate, Steer, and Improve: A Practical Survey of Actionable Mechanistic Interpretability in Large Language Models","ref_index":133,"is_internal_anchor":false},{"citing_arxiv_id":"2404.18930","citing_title":"Hallucination of Multimodal Large Language Models: A Survey","ref_index":80,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12582","citing_title":"Relaxing Anchor-Frame Dominance for Mitigating Hallucinations in Video Large Language Models","ref_index":15,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OHACGJCNDB4YAPGULQ4US4BBVV","json":"https://pith.science/pith/OHACGJCNDB4YAPGULQ4US4BBVV.json","graph_json":"https://pith.science/api/pith-number/OHACGJCNDB4YAPGULQ4US4BBVV/graph.json","events_json":"https://pith.science/api/pith-number/OHACGJCNDB4YAPGULQ4US4BBVV/events.json","paper":"https://pith.science/paper/OHACGJCN"},"agent_actions":{"view_html":"https://pith.science/pith/OHACGJCNDB4YAPGULQ4US4BBVV","download_json":"https://pith.science/pith/OHACGJCNDB4YAPGULQ4US4BBVV.json","view_paper":"https://pith.science/paper/OHACGJCN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.02762&json=true","fetch_graph":"https://pith.science/api/pith-number/OHACGJCNDB4YAPGULQ4US4BBVV/graph.json","fetch_events":"https://pith.science/api/pith-number/OHACGJCNDB4YAPGULQ4US4BBVV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OHACGJCNDB4YAPGULQ4US4BBVV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OHACGJCNDB4YAPGULQ4US4BBVV/action/storage_attestation","attest_author":"https://pith.science/pith/OHACGJCNDB4YAPGULQ4US4BBVV/action/author_attestation","sign_citation":"https://pith.science/pith/OHACGJCNDB4YAPGULQ4US4BBVV/action/citation_signature","submit_replication":"https://pith.science/pith/OHACGJCNDB4YAPGULQ4US4BBVV/action/replication_record"}},"created_at":"2026-07-05T10:12:20.856828+00:00","updated_at":"2026-07-05T10:12:20.856828+00:00"}