{"id":6232,"date":"2025-02-12T05:50:00","date_gmt":"2025-02-11T20:50:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6232"},"modified":"2025-02-12T05:50:00","modified_gmt":"2025-02-11T20:50:00","slug":"teaching-language-models-to-critique-via-reinforcement-learning","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6232","title":{"rendered":"Teaching Language Models to Critique via Reinforcement Learning\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Teaching Language Models to Critique via Reinforcement Learning\u00a0<\/strong>[59.4]<br>\u6211\u3005\u306f\u3001CTRL\u3067\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3055\u308c\u305f\u6279\u8a55\u5bb6\u304c\u3001\u30d1\u30b9\u30ec\u30fc\u30c8\u3092\u5927\u5e45\u306b\u5411\u4e0a\u3057\u3001\u30d9\u30fc\u30b9\u30e2\u30c7\u30eb\u3068\u3088\u308a\u5f37\u529b\u306a\u30b8\u30a7\u30cd\u30ec\u30fc\u30bf\u30e2\u30c7\u30eb\u306e\u4e21\u65b9\u3067\u30a8\u30e9\u30fc\u3092\u8efd\u6e1b\u3059\u308b\u3053\u3068\u3092\u793a\u3057\u305f\u3002 \u307e\u305f\u3001\u3053\u308c\u3089\u306e\u6279\u5224\u30e2\u30c7\u30eb\u304c\u6b63\u78ba\u306a\u751f\u6210\u5831\u916c\u30e2\u30c7\u30eb\u3068\u3057\u3066\u6a5f\u80fd\u3057\u3001\u53cd\u5fa9\u7684\u6279\u8a55\u30fb\u4fee\u6b63\u306b\u3088\u308b\u30c6\u30b9\u30c8\u30bf\u30a4\u30e0\u30b9\u30b1\u30fc\u30ea\u30f3\u30b0\u3092\u53ef\u80fd\u306b\u3059\u308b\u3053\u3068\u3092\u793a\u3059\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2502.03492v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2502.03492v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Wed, 05 Feb 2025 02:18:46 GMT)<\/li>\n\n\n\n<li>\u300ctwo-stage training approach: (1) synthesizing high-quality critiques by reasoning about execution feedback, then (2) refining the critic through reinforcement learning.\u300d\u3068\u3044\u30462\u30b9\u30c6\u30fc\u30b8\u69cb\u6210\u3001\u5f37\u5316\u5b66\u7fd2\uff08GRPO\uff09\u3092\u6d3b\u7528\u3057\u305fcritic\u30e2\u30c7\u30eb\u306e\u69cb\u7bc9\u3002<\/li>\n\n\n\n<li>\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u30b5\u30a4\u30c8\u306f<a href=\"https:\/\/critic-rl.github.io\/\">CTRL: Critic Training via Reinforcement Learning<\/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":[84,576],"class_list":["post-6232","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-critic","tag-576"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6232","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=6232"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6232\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6232"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6232"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6232"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}