{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:YQKWMRID7BLHSJ4GTHWJMY65F7","short_pith_number":"pith:YQKWMRID","schema_version":"1.0","canonical_sha256":"c415664503f85679278699ec9663dd2fed634893681b4c1f3e00df1979e93f8d","source":{"kind":"arxiv","id":"2506.07936","version":1},"attestation_state":"computed","paper":{"title":"Mimicking or Reasoning: Rethinking Multi-Modal In-Context Learning in Vision-Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Chengyue Huang, Jingyun Xiao, Moises Andrade, Shivang Chopra, Sichen Zhu, Yuchen Zhu, Zsolt Kira","submitted_at":"2025-06-09T16:55:32Z","abstract_excerpt":"Vision-language models (VLMs) are widely assumed to exhibit in-context learning (ICL), a property similar to that of their language-only counterparts. While recent work suggests VLMs can perform multimodal ICL (MM-ICL), studies show they often rely on shallow heuristics -- such as copying or majority voting -- rather than true task understanding. We revisit this assumption by evaluating VLMs under distribution shifts, where support examples come from a dataset different from the query. Surprisingly, performance often degrades with more demonstrations, and models tend to copy answers rather tha"},"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":"2506.07936","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-06-09T16:55:32Z","cross_cats_sorted":["cs.AI","cs.CL","cs.LG"],"title_canon_sha256":"9f237381de49ff9065e8db1921f758f891d2e481d4194623df0b133c3038bfda","abstract_canon_sha256":"6df3d42b66e56170f19c828535c05731e32753d1df7b79b7f7609053966849a0"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:18:35.687462Z","signature_b64":"A6VZOvIdFs3xflkb6/dqLXNpagIMd69jeHmECWSxgalDvUB8eedNO5f5m2GDFHUAbISlZKvimsktCyZeQf3iAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c415664503f85679278699ec9663dd2fed634893681b4c1f3e00df1979e93f8d","last_reissued_at":"2026-07-05T11:18:35.687004Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:18:35.687004Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mimicking or Reasoning: Rethinking Multi-Modal In-Context Learning in Vision-Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Chengyue Huang, Jingyun Xiao, Moises Andrade, Shivang Chopra, Sichen Zhu, Yuchen Zhu, Zsolt Kira","submitted_at":"2025-06-09T16:55:32Z","abstract_excerpt":"Vision-language models (VLMs) are widely assumed to exhibit in-context learning (ICL), a property similar to that of their language-only counterparts. While recent work suggests VLMs can perform multimodal ICL (MM-ICL), studies show they often rely on shallow heuristics -- such as copying or majority voting -- rather than true task understanding. We revisit this assumption by evaluating VLMs under distribution shifts, where support examples come from a dataset different from the query. Surprisingly, performance often degrades with more demonstrations, and models tend to copy answers rather tha"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.07936","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/2506.07936/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":"2506.07936","created_at":"2026-07-05T11:18:35.687073+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.07936v1","created_at":"2026-07-05T11:18:35.687073+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.07936","created_at":"2026-07-05T11:18:35.687073+00:00"},{"alias_kind":"pith_short_12","alias_value":"YQKWMRID7BLH","created_at":"2026-07-05T11:18:35.687073+00:00"},{"alias_kind":"pith_short_16","alias_value":"YQKWMRID7BLHSJ4G","created_at":"2026-07-05T11:18:35.687073+00:00"},{"alias_kind":"pith_short_8","alias_value":"YQKWMRID","created_at":"2026-07-05T11:18:35.687073+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2509.18095","citing_title":"MetaEmbed: Scaling Multimodal Retrieval at Test-Time with Flexible Late Interaction","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01333","citing_title":"OralMLLM-Bench: Evaluating Cognitive Capabilities of Multimodal Large Language Models in Dental Practice","ref_index":44,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13403","citing_title":"Why Multimodal In-Context Learning Lags Behind? Unveiling the Inner Mechanisms and Bottlenecks","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YQKWMRID7BLHSJ4GTHWJMY65F7","json":"https://pith.science/pith/YQKWMRID7BLHSJ4GTHWJMY65F7.json","graph_json":"https://pith.science/api/pith-number/YQKWMRID7BLHSJ4GTHWJMY65F7/graph.json","events_json":"https://pith.science/api/pith-number/YQKWMRID7BLHSJ4GTHWJMY65F7/events.json","paper":"https://pith.science/paper/YQKWMRID"},"agent_actions":{"view_html":"https://pith.science/pith/YQKWMRID7BLHSJ4GTHWJMY65F7","download_json":"https://pith.science/pith/YQKWMRID7BLHSJ4GTHWJMY65F7.json","view_paper":"https://pith.science/paper/YQKWMRID","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.07936&json=true","fetch_graph":"https://pith.science/api/pith-number/YQKWMRID7BLHSJ4GTHWJMY65F7/graph.json","fetch_events":"https://pith.science/api/pith-number/YQKWMRID7BLHSJ4GTHWJMY65F7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YQKWMRID7BLHSJ4GTHWJMY65F7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YQKWMRID7BLHSJ4GTHWJMY65F7/action/storage_attestation","attest_author":"https://pith.science/pith/YQKWMRID7BLHSJ4GTHWJMY65F7/action/author_attestation","sign_citation":"https://pith.science/pith/YQKWMRID7BLHSJ4GTHWJMY65F7/action/citation_signature","submit_replication":"https://pith.science/pith/YQKWMRID7BLHSJ4GTHWJMY65F7/action/replication_record"}},"created_at":"2026-07-05T11:18:35.687073+00:00","updated_at":"2026-07-05T11:18:35.687073+00:00"}