{"id":8359,"date":"2026-03-26T05:22:00","date_gmt":"2026-03-25T20:22:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=8359"},"modified":"2026-03-21T15:27:36","modified_gmt":"2026-03-21T06:27:36","slug":"cognitive-mismatch-in-multimodal-large-language-models-for-discrete-symbol-understanding","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=8359","title":{"rendered":"Cognitive Mismatch in Multimodal Large Language Models for Discrete Symbol Understanding\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Cognitive Mismatch in Multimodal Large Language Models for Discrete Symbol Understanding\u00a0<\/strong>[96.8]<br>\u672c\u7a3f\u3067\u306f,\u6700\u4e0a\u4f4d\u306eMLLM\u304c\u500b\u5225\u306e\u610f\u5473\u7a7a\u9593\u3092\u3069\u306e\u3088\u3046\u306b\u30ca\u30d3\u30b2\u30fc\u30c8\u3059\u308b\u304b\u3092\u8a55\u4fa1\u3059\u308b\u305f\u3081\u306e\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3092\u7d39\u4ecb\u3059\u308b\u3002 \u30e2\u30c7\u30eb\u306f\u57fa\u672c\u7684\u306a\u30b7\u30f3\u30dc\u30eb\u8a8d\u8b58\u306b\u5931\u6557\u3059\u308b\u3053\u3068\u304c\u591a\u3044\u304c\u3001\u8907\u96d1\u306a\u63a8\u8ad6\u30bf\u30b9\u30af\u306b\u6210\u529f\u3057\u3066\u3044\u308b\u3002 \u3053\u306e\u4f5c\u696d\u306f\u3001\u3088\u308a\u53b3\u683c\u3067\u4eba\u9593\u6307\u5411\u306e\u30a4\u30f3\u30c6\u30ea\u30b8\u30a7\u30f3\u30c8\u306a\u30b7\u30b9\u30c6\u30e0\u3092\u958b\u767a\u3059\u308b\u305f\u3081\u306e\u30ed\u30fc\u30c9\u30de\u30c3\u30d7\u3092\u63d0\u4f9b\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2603.18472v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2603.18472v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 19 Mar 2026 04:08:20 GMT)<\/li>\n\n\n\n<li>\u300cdespite impressive reasoning capabilities, current models frequently fail at foundational visual symbol grounding, relying instead on linguistic priors, procedural imitation, or memorized patterns. Our findings challenge a prevailing assumption in multimodal intelligence that visual recognition is inherently simpler than reasoning. Instead, we observe a consistent recognition-reasoning inversion phenomenon, where higher-level reasoning performance often masks deficiencies in low-level symbolic perception. This phenomenon underscores a key limitation of existing training paradigms: while models excel at leveraging large-scale continual natural images, they struggle to construct stable, compositional visual representations of abstract, discrete symbols.\u300d\u3068\u3044\u3046\u9762\u767d\u3044\u6307\u6458\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[517],"class_list":["post-8359","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-517"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8359","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=8359"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8359\/revisions"}],"predecessor-version":[{"id":8360,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8359\/revisions\/8360"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8359"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8359"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8359"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}