{"id":8361,"date":"2026-03-25T03:28:00","date_gmt":"2026-03-24T18:28:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=8361"},"modified":"2026-03-21T15:33:37","modified_gmt":"2026-03-21T06:33:37","slug":"decoding-the-critique-mechanism-in-large-reasoning-models","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=8361","title":{"rendered":"Decoding the Critique Mechanism in Large Reasoning Models"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Decoding the Critique Mechanism in Large Reasoning Models&nbsp;<\/strong>[50.8]<br>\u5927\u898f\u6a21\u63a8\u8ad6\u30e2\u30c7\u30eb(LRM)\u306f\u3001\u30d0\u30c3\u30af\u30c8\u30e9\u30c3\u30af\u3068\u81ea\u5df1\u691c\u8a3c\u30e1\u30ab\u30cb\u30ba\u30e0\u3092\u793a\u3057\u3001\u4e2d\u9593\u30b9\u30c6\u30c3\u30d7\u3092\u4fee\u6b63\u3057\u3066\u6b63\u3057\u3044\u89e3\u306b\u5230\u9054\u3067\u304d\u308b\u3088\u3046\u306b\u3059\u308b\u3002 \u4e2d\u9593\u63a8\u8ad6\u30b9\u30c6\u30c3\u30d7\u306b\u7b97\u8853\u30df\u30b9\u3092\u633f\u5165\u3059\u308b\u3053\u3068\u306b\u3088\u308a,\u73fe\u5728\u306eLEM\u304c\u30a8\u30e9\u30fc\u304b\u3089\u3069\u306e\u3088\u3046\u306b\u56de\u5fa9\u3059\u308b\u304b\u3092\u691c\u8a0e\u3059\u308b\u3002 \u30c1\u30a7\u30fc\u30f3\u30fb\u30aa\u30d6\u30fb\u30b7\u30fc\u30af\u30ec\u30c3\u30c8\u3092\u4f1d\u64ad\u3059\u308b\u8aa4\u308a\u306b\u3082\u304b\u304b\u308f\u3089\u305a\u3001\u30e2\u30c7\u30eb\u306f\u4f9d\u7136\u3068\u3057\u3066\u6b63\u3057\u3044\u6700\u7d42\u89e3\u306b\u9054\u3057\u3066\u3044\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2603.16331v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2603.16331v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Tue, 17 Mar 2026 10:03:30 GMT)<\/li>\n\n\n\n<li>\u300cwe demonstrate that the critique vector influences test-time scaling: increasing it helps the model better recognize its mistakes and improve its final accuracy, while decreasing it degrades performance.\u300d\u3068\u3044\u3046\u306e\u304c\u3068\u3066\u3082\u8208\u5473\u6df1\u3044\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/mail-research\/lrm-critique-vectors\">GitHub &#8211; mail-research\/lrm-critique-vectors \u00b7 GitHub<\/a><\/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":[84],"class_list":["post-8361","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-critic"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8361","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=8361"}],"version-history":[{"count":2,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8361\/revisions"}],"predecessor-version":[{"id":8363,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8361\/revisions\/8363"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8361"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8361"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8361"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}