{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SSRMBP7DUQPRO3AOWNMVPSBRJ3","short_pith_number":"pith:SSRMBP7D","schema_version":"1.0","canonical_sha256":"94a2c0bfe3a41f176c0eb35957c8314ee9d8ec09148fa875eb1803c19614f0a4","source":{"kind":"arxiv","id":"2607.08059","version":1},"attestation_state":"computed","paper":{"title":"When Thinking Hurts: Epistemic Signals in the Reasoning Chains of Visual Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Mayank Singal","submitted_at":"2026-07-09T02:19:16Z","abstract_excerpt":"Uncertainty quantification for visual language models (VLMs) conventionally targets the answer token distribution. We provide the first three-family empirical characterisation of answer entropy behaviour in thinking-mode VLMs. Running four models on identical POPE adversarial samples, we find three qualitatively distinct patterns: Qwen3-VL-8B-Thinking shows complete collapse (ans H AUROC = 0.492); GLM-4.1V-9B-Thinking shows no collapse (0.716); and InternVL3-8B shows selective thinking (chains on only 50% of samples, ans H = 0.675 full / 0.602 thinking-only). Across all three thinking-mode mod"},"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":"2607.08059","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-09T02:19:16Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"f877aff6fb4822fc6aeb611ddee007a15c6d5ca8cdca2c536d0107a36e3c2f55","abstract_canon_sha256":"471adb48261e858ff0840a47c3e2306c6e1581a5fcac127b023198725cd43759"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-10T00:19:09.262999Z","signature_b64":"Km7lXW7MfZ0bGM/NzQMe92rgxDIbRA4wkiEVTdCdGqx4gRVh2Aiuec0OxRqvJMwcJZznNrrvb5dIfF1tl3sPAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"94a2c0bfe3a41f176c0eb35957c8314ee9d8ec09148fa875eb1803c19614f0a4","last_reissued_at":"2026-07-10T00:19:09.262598Z","signature_status":"signed_v1","first_computed_at":"2026-07-10T00:19:09.262598Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Thinking Hurts: Epistemic Signals in the Reasoning Chains of Visual Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Mayank Singal","submitted_at":"2026-07-09T02:19:16Z","abstract_excerpt":"Uncertainty quantification for visual language models (VLMs) conventionally targets the answer token distribution. We provide the first three-family empirical characterisation of answer entropy behaviour in thinking-mode VLMs. Running four models on identical POPE adversarial samples, we find three qualitatively distinct patterns: Qwen3-VL-8B-Thinking shows complete collapse (ans H AUROC = 0.492); GLM-4.1V-9B-Thinking shows no collapse (0.716); and InternVL3-8B shows selective thinking (chains on only 50% of samples, ans H = 0.675 full / 0.602 thinking-only). Across all three thinking-mode mod"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.08059","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/2607.08059/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":"2607.08059","created_at":"2026-07-10T00:19:09.262662+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.08059v1","created_at":"2026-07-10T00:19:09.262662+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.08059","created_at":"2026-07-10T00:19:09.262662+00:00"},{"alias_kind":"pith_short_12","alias_value":"SSRMBP7DUQPR","created_at":"2026-07-10T00:19:09.262662+00:00"},{"alias_kind":"pith_short_16","alias_value":"SSRMBP7DUQPRO3AO","created_at":"2026-07-10T00:19:09.262662+00:00"},{"alias_kind":"pith_short_8","alias_value":"SSRMBP7D","created_at":"2026-07-10T00:19:09.262662+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SSRMBP7DUQPRO3AOWNMVPSBRJ3","json":"https://pith.science/pith/SSRMBP7DUQPRO3AOWNMVPSBRJ3.json","graph_json":"https://pith.science/api/pith-number/SSRMBP7DUQPRO3AOWNMVPSBRJ3/graph.json","events_json":"https://pith.science/api/pith-number/SSRMBP7DUQPRO3AOWNMVPSBRJ3/events.json","paper":"https://pith.science/paper/SSRMBP7D"},"agent_actions":{"view_html":"https://pith.science/pith/SSRMBP7DUQPRO3AOWNMVPSBRJ3","download_json":"https://pith.science/pith/SSRMBP7DUQPRO3AOWNMVPSBRJ3.json","view_paper":"https://pith.science/paper/SSRMBP7D","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.08059&json=true","fetch_graph":"https://pith.science/api/pith-number/SSRMBP7DUQPRO3AOWNMVPSBRJ3/graph.json","fetch_events":"https://pith.science/api/pith-number/SSRMBP7DUQPRO3AOWNMVPSBRJ3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SSRMBP7DUQPRO3AOWNMVPSBRJ3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SSRMBP7DUQPRO3AOWNMVPSBRJ3/action/storage_attestation","attest_author":"https://pith.science/pith/SSRMBP7DUQPRO3AOWNMVPSBRJ3/action/author_attestation","sign_citation":"https://pith.science/pith/SSRMBP7DUQPRO3AOWNMVPSBRJ3/action/citation_signature","submit_replication":"https://pith.science/pith/SSRMBP7DUQPRO3AOWNMVPSBRJ3/action/replication_record"}},"created_at":"2026-07-10T00:19:09.262662+00:00","updated_at":"2026-07-10T00:19:09.262662+00:00"}