{"id":8224,"date":"2026-02-18T06:09:00","date_gmt":"2026-02-17T21:09:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=8224"},"modified":"2026-02-14T22:12:21","modified_gmt":"2026-02-14T13:12:21","slug":"outcome-accuracy-is-not-enough-aligning-the-reasoning-process-of-reward-models","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=8224","title":{"rendered":"Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Outcome Accuracy is Not Enough: Aligning the Reasoning Process of Reward Models\u00a0<\/strong>[108.3]<br>Rationale Consistency\u306f\u3001\u30e2\u30c7\u30eb\u306e\u63a8\u8ad6\u30d7\u30ed\u30bb\u30b9\u3068\u4eba\u9593\u306e\u5224\u65ad\u306e\u30a2\u30e9\u30a4\u30e1\u30f3\u30c8\u3092\u5b9a\u91cf\u5316\u3059\u308b\u3001\u304d\u3081\u7d30\u304b\u3044\u8a08\u91cf\u3067\u3042\u308b\u3002 \u6211\u3005\u306e\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2\u30e2\u30c7\u30eb\u306e\u8a55\u4fa1\u3067\u306f,\u6700\u5148\u7aef\u30e2\u30c7\u30eb\u9593\u3067\u5408\u7406\u7684\u306a\u4e00\u8cab\u6027\u304c\u52b9\u679c\u7684\u306b\u8b58\u5225\u3067\u304d\u308b\u3053\u3068\u304c\u793a\u3055\u308c\u3066\u3044\u308b\u3002 \u6211\u3005\u306f\u3001GenRM\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u306e\u5408\u7406\u6027\u4e00\u8cab\u6027\u3068\u7d50\u679c\u7cbe\u5ea6\u3092\u7d44\u307f\u5408\u308f\u305b\u305f\u30cf\u30a4\u30d6\u30ea\u30c3\u30c9\u4fe1\u53f7\u3092\u5c0e\u5165\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2602.04649v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2602.04649v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Wed, 04 Feb 2026 15:24:52 GMT)<\/li>\n\n\n\n<li>\u300cOur evaluation of frontier models reveals that rationale consistency effectively discriminates among state-of-the-art models and detects deceptive alignment, while outcome accuracy falls short in both respects. To mitigate this gap, we introduce a hybrid signal that combines rationale consistency with outcome accuracy for GenRM training.\u300d\u3068\u306e\u3053\u3068\u3002\u524d\u534a\u306e\u6307\u6458\u306b\u3064\u3044\u3066\u76f4\u89b3\u7684\u306b\u306f\u305d\u3046\u3060\u3068\u601d\u3046\u3082\u306e\u306e\u8208\u5473\u6df1\u3044\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/QwenLM\/RationaleRM\">GitHub &#8211; QwenLM\/RationaleRM<\/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":[576],"class_list":["post-8224","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-576"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8224","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=8224"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8224\/revisions"}],"predecessor-version":[{"id":8225,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8224\/revisions\/8225"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8224"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8224"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8224"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}