{"id":6682,"date":"2025-05-06T07:38:00","date_gmt":"2025-05-05T22:38:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6682"},"modified":"2025-05-06T07:38:00","modified_gmt":"2025-05-05T22:38:00","slug":"spc-evolving-self-play-critic-via-adversarial-games-for-llm-reasoning","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6682","title":{"rendered":"SPC: Evolving Self-Play Critic via Adversarial Games for LLM Reasoning"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>SPC: Evolving Self-Play Critic via Adversarial Games for LLM Reasoning\u00a0<\/strong>[99.6]<br>\u30bb\u30eb\u30d5\u30d7\u30ec\u30a4\u6279\u5224(Self-Play Critic\u3001SPC)\u306f\u3001\u5bfe\u6226\u578b\u30bb\u30eb\u30d5\u30d7\u30ec\u30a4\u30b2\u30fc\u30e0\u3092\u901a\u3058\u3066\u63a8\u8ad6\u30b9\u30c6\u30c3\u30d7\u3092\u8a55\u4fa1\u3059\u308b\u80fd\u529b\u3092\u9032\u5316\u3055\u305b\u308b\u65b0\u3057\u3044\u30a2\u30d7\u30ed\u30fc\u30c1\u3067\u3042\u308b\u3002 SPC\u306f\u3001\u30d9\u30fc\u30b9\u30e2\u30c7\u30eb\u306e2\u3064\u306e\u30b3\u30d4\u30fc\u3092\u5fae\u8abf\u6574\u3057\u3066\u30012\u3064\u306e\u5f79\u5272\u3001\u3059\u306a\u308f\u3061\u300c\u30b9\u30cb\u30fc\u30ad\u30fc\u30b8\u30a7\u30cd\u30ec\u30fc\u30bf\u300d\u3068\u300c\u6279\u5224\u7684\u300d\u3092\u6f14\u3058\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2504.19162v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2504.19162v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Sun, 27 Apr 2025 08:45:06 GMT)<\/li>\n\n\n\n<li>\u300cIn this paper, we propose a self-play critic with the ability of detecting step-level LLMs reasoning errors. Specifically, we design a sneaky generator to produce incorrect steps and a critic to assess the correctness of each step. Through the adversarial game between these two models, we can continuously generate positive and negative samples for reinforcement learning.\u300d\u3068\u3044\u3046\u30a2\u30d7\u30ed\u30fc\u30c1\u306e\u63d0\u6848\u3002GAN\u3063\u307d\u3044\u306a\u3068\u601d\u3046\u3002<\/li>\n\n\n\n<li>\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u30b5\u30a4\u30c8\u306f<a href=\"https:\/\/chen-judge.github.io\/SPC\/\">SPC: Evolving Self-Play Critic via Adversarial Games for LLM Reasoning<\/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":[84,356],"class_list":["post-6682","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-critic","tag-self-x"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6682","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=6682"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6682\/revisions"}],"predecessor-version":[{"id":6703,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6682\/revisions\/6703"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6682"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6682"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6682"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}