{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7D6UCQRCAW5NHWKV2GPQ7PB6MD","short_pith_number":"pith:7D6UCQRC","schema_version":"1.0","canonical_sha256":"f8fd41422205bad3d955d19f0fbc3e60d23852d4b0154b35bb8ddbf147837a65","source":{"kind":"arxiv","id":"2506.08008","version":1},"attestation_state":"computed","paper":{"title":"Hidden in plain sight: VLMs overlook their visual representations","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Devin Guillory, Stephanie Fu, Trevor Darrell, Tyler Bonnen","submitted_at":"2025-06-09T17:59:54Z","abstract_excerpt":"Language provides a natural interface to specify and evaluate performance on visual tasks. To realize this possibility, vision language models (VLMs) must successfully integrate visual and linguistic information. Our work compares VLMs to a direct readout of their visual encoders to understand their ability to integrate across these modalities. Across a series of vision-centric benchmarks (e.g., depth estimation, correspondence), we find that VLMs perform substantially worse than their visual encoders, dropping to near-chance performance. We investigate these results through a series of analys"},"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":"2506.08008","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CV","submitted_at":"2025-06-09T17:59:54Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"c9060cd254f77d0d761d1d857b86003cdc834d66579c2b1a6b61c0f51e1a4253","abstract_canon_sha256":"79b38f8fffb8eca7d934b6bada7881ba908b44998079df374ce2edee39da57d4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:18:37.185261Z","signature_b64":"P7kqk1DcPD1J4tW3llv+WMN0DU2AcQMwzOIdIXzMYq5+Qx3j1Bm6xuFrApPBQMGoX//FZiysPTz7zi+msr07BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f8fd41422205bad3d955d19f0fbc3e60d23852d4b0154b35bb8ddbf147837a65","last_reissued_at":"2026-07-05T11:18:37.184781Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:18:37.184781Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Hidden in plain sight: VLMs overlook their visual representations","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Devin Guillory, Stephanie Fu, Trevor Darrell, Tyler Bonnen","submitted_at":"2025-06-09T17:59:54Z","abstract_excerpt":"Language provides a natural interface to specify and evaluate performance on visual tasks. To realize this possibility, vision language models (VLMs) must successfully integrate visual and linguistic information. Our work compares VLMs to a direct readout of their visual encoders to understand their ability to integrate across these modalities. Across a series of vision-centric benchmarks (e.g., depth estimation, correspondence), we find that VLMs perform substantially worse than their visual encoders, dropping to near-chance performance. We investigate these results through a series of analys"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.08008","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/2506.08008/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":"2506.08008","created_at":"2026-07-05T11:18:37.184838+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.08008v1","created_at":"2026-07-05T11:18:37.184838+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.08008","created_at":"2026-07-05T11:18:37.184838+00:00"},{"alias_kind":"pith_short_12","alias_value":"7D6UCQRCAW5N","created_at":"2026-07-05T11:18:37.184838+00:00"},{"alias_kind":"pith_short_16","alias_value":"7D6UCQRCAW5NHWKV","created_at":"2026-07-05T11:18:37.184838+00:00"},{"alias_kind":"pith_short_8","alias_value":"7D6UCQRC","created_at":"2026-07-05T11:18:37.184838+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":14,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.20077","citing_title":"The Hidden Evolution of Disguised Visual Context inside the VLM","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2606.06890","citing_title":"Diagnosing Visual Ignorance in Vision-Language Models","ref_index":39,"is_internal_anchor":false},{"citing_arxiv_id":"2510.16416","citing_title":"SSL4RL: Revisiting Self-supervised Learning as Intrinsic Reward for Visual-Language Reasoning","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2603.02667","citing_title":"Unifying Contrastive and Generative Objectives for Visual Understanding and Text-to-Image Generation","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13161","citing_title":"A$_3$B$_2$: Adaptive Asymmetric Adapter for Alleviating Branch Bias in Vision-Language Image Classification with Few-Shot Learning","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2512.10941","citing_title":"Mull-Tokens: Modality-Agnostic Latent Thinking","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13161","citing_title":"A$_3$B$_2$: Adaptive Asymmetric Adapter for Alleviating Branch Bias in Vision-Language Image Classification with Few-Shot Learning","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.02486","citing_title":"VLMs Need Words: Vision Language Models Ignore Visual Detail In Favor of Semantic Anchors","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09693","citing_title":"Do multimodal models imagine electric sheep?","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2604.12966","citing_title":"Boosting Visual Instruction Tuning with Self-Supervised Guidance","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08333","citing_title":"Lost in the Hype: Revealing and Dissecting the Performance Degradation of Medical Multimodal Large Language Models in Image Classification","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09425","citing_title":"Do Vision Language Models Need to Process Image Tokens?","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05117","citing_title":"Watch Before You Answer: Learning from Visually Grounded Post-Training","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2604.14363","citing_title":"The Cost of Language: Centroid Erasure Exposes and Exploits Modal Competition in Multimodal Language Models","ref_index":4,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7D6UCQRCAW5NHWKV2GPQ7PB6MD","json":"https://pith.science/pith/7D6UCQRCAW5NHWKV2GPQ7PB6MD.json","graph_json":"https://pith.science/api/pith-number/7D6UCQRCAW5NHWKV2GPQ7PB6MD/graph.json","events_json":"https://pith.science/api/pith-number/7D6UCQRCAW5NHWKV2GPQ7PB6MD/events.json","paper":"https://pith.science/paper/7D6UCQRC"},"agent_actions":{"view_html":"https://pith.science/pith/7D6UCQRCAW5NHWKV2GPQ7PB6MD","download_json":"https://pith.science/pith/7D6UCQRCAW5NHWKV2GPQ7PB6MD.json","view_paper":"https://pith.science/paper/7D6UCQRC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.08008&json=true","fetch_graph":"https://pith.science/api/pith-number/7D6UCQRCAW5NHWKV2GPQ7PB6MD/graph.json","fetch_events":"https://pith.science/api/pith-number/7D6UCQRCAW5NHWKV2GPQ7PB6MD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7D6UCQRCAW5NHWKV2GPQ7PB6MD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7D6UCQRCAW5NHWKV2GPQ7PB6MD/action/storage_attestation","attest_author":"https://pith.science/pith/7D6UCQRCAW5NHWKV2GPQ7PB6MD/action/author_attestation","sign_citation":"https://pith.science/pith/7D6UCQRCAW5NHWKV2GPQ7PB6MD/action/citation_signature","submit_replication":"https://pith.science/pith/7D6UCQRCAW5NHWKV2GPQ7PB6MD/action/replication_record"}},"created_at":"2026-07-05T11:18:37.184838+00:00","updated_at":"2026-07-05T11:18:37.184838+00:00"}