{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:4QWMCD4KXZB6IFUOCBLB5DJXR2","short_pith_number":"pith:4QWMCD4K","schema_version":"1.0","canonical_sha256":"e42cc10f8abe43e4168e10561e8d378eb7c7f854a541cb3bb994552b7c2a6326","source":{"kind":"arxiv","id":"2403.09037","version":2},"attestation_state":"computed","paper":{"title":"The First to Know: How Token Distributions Reveal Hidden Knowledge in Large Vision-Language Models?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Akshay Asthana, Kartik Gupta, Liang Zheng, Ming Xu, Qinyu Zhao, Stephen Gould","submitted_at":"2024-03-14T02:25:35Z","abstract_excerpt":"Large vision-language models (LVLMs), designed to interpret and respond to human instructions, occasionally generate hallucinated or harmful content due to inappropriate instructions. This study uses linear probing to shed light on the hidden knowledge at the output layers of LVLMs. We demonstrate that the logit distributions of the first tokens contain sufficient information to determine whether to respond to the instructions, including recognizing unanswerable visual questions, defending against jailbreaking attacks, and identifying deceptive questions. Such hidden knowledge is gradually los"},"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":"2403.09037","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2024-03-14T02:25:35Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"75230d65a6e59622f21403ff8b99b9d0de8812b6eee3dfafa370ab2792321835","abstract_canon_sha256":"0c441287cf35e6d8d4e4301adc22ccf23aba3fc64f712288c9c038b395045753"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:44:51.889073Z","signature_b64":"sowgG+Z0Y0fMov5KGkovNaddgVfswJxRkSpiOML25qu6JSG6xlwKeqsRebk/Ba5uxNRH9EXVI6HOJ9ZQAeBjCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e42cc10f8abe43e4168e10561e8d378eb7c7f854a541cb3bb994552b7c2a6326","last_reissued_at":"2026-07-05T08:44:51.888658Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:44:51.888658Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The First to Know: How Token Distributions Reveal Hidden Knowledge in Large Vision-Language Models?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Akshay Asthana, Kartik Gupta, Liang Zheng, Ming Xu, Qinyu Zhao, Stephen Gould","submitted_at":"2024-03-14T02:25:35Z","abstract_excerpt":"Large vision-language models (LVLMs), designed to interpret and respond to human instructions, occasionally generate hallucinated or harmful content due to inappropriate instructions. This study uses linear probing to shed light on the hidden knowledge at the output layers of LVLMs. We demonstrate that the logit distributions of the first tokens contain sufficient information to determine whether to respond to the instructions, including recognizing unanswerable visual questions, defending against jailbreaking attacks, and identifying deceptive questions. Such hidden knowledge is gradually los"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2403.09037","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/2403.09037/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":"2403.09037","created_at":"2026-07-05T08:44:51.888714+00:00"},{"alias_kind":"arxiv_version","alias_value":"2403.09037v2","created_at":"2026-07-05T08:44:51.888714+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2403.09037","created_at":"2026-07-05T08:44:51.888714+00:00"},{"alias_kind":"pith_short_12","alias_value":"4QWMCD4KXZB6","created_at":"2026-07-05T08:44:51.888714+00:00"},{"alias_kind":"pith_short_16","alias_value":"4QWMCD4KXZB6IFUO","created_at":"2026-07-05T08:44:51.888714+00:00"},{"alias_kind":"pith_short_8","alias_value":"4QWMCD4K","created_at":"2026-07-05T08:44:51.888714+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.22007","citing_title":"Hallucination as Commitment Failure: Larger LLMs Misfire Despite Knowing the Answer","ref_index":61,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4QWMCD4KXZB6IFUOCBLB5DJXR2","json":"https://pith.science/pith/4QWMCD4KXZB6IFUOCBLB5DJXR2.json","graph_json":"https://pith.science/api/pith-number/4QWMCD4KXZB6IFUOCBLB5DJXR2/graph.json","events_json":"https://pith.science/api/pith-number/4QWMCD4KXZB6IFUOCBLB5DJXR2/events.json","paper":"https://pith.science/paper/4QWMCD4K"},"agent_actions":{"view_html":"https://pith.science/pith/4QWMCD4KXZB6IFUOCBLB5DJXR2","download_json":"https://pith.science/pith/4QWMCD4KXZB6IFUOCBLB5DJXR2.json","view_paper":"https://pith.science/paper/4QWMCD4K","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2403.09037&json=true","fetch_graph":"https://pith.science/api/pith-number/4QWMCD4KXZB6IFUOCBLB5DJXR2/graph.json","fetch_events":"https://pith.science/api/pith-number/4QWMCD4KXZB6IFUOCBLB5DJXR2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4QWMCD4KXZB6IFUOCBLB5DJXR2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4QWMCD4KXZB6IFUOCBLB5DJXR2/action/storage_attestation","attest_author":"https://pith.science/pith/4QWMCD4KXZB6IFUOCBLB5DJXR2/action/author_attestation","sign_citation":"https://pith.science/pith/4QWMCD4KXZB6IFUOCBLB5DJXR2/action/citation_signature","submit_replication":"https://pith.science/pith/4QWMCD4KXZB6IFUOCBLB5DJXR2/action/replication_record"}},"created_at":"2026-07-05T08:44:51.888714+00:00","updated_at":"2026-07-05T08:44:51.888714+00:00"}