{"id":6885,"date":"2025-06-11T05:06:00","date_gmt":"2025-06-10T20:06:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6885"},"modified":"2025-06-07T21:12:34","modified_gmt":"2025-06-07T12:12:34","slug":"self-challenging-language-model-agents","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6885","title":{"rendered":"Self-Challenging Language Model Agents"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Self-Challenging Language Model Agents\u00a0<\/strong>[98.6]<br>\u672c\u7a3f\u3067\u306f,\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u304c\u81ea\u3089\u751f\u6210\u3059\u308b\u9ad8\u54c1\u8cea\u306a\u30bf\u30b9\u30af\u306b\u3064\u3044\u3066,\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u3092\u8a13\u7df4\u3059\u308b\u305f\u3081\u306e\u30bb\u30eb\u30d5\u30c1\u30a7\u30f3\u30b8\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3092\u63d0\u6848\u3059\u308b\u3002 \u3053\u306e\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306f\u3001Llama-3.1-8B-Instruct\u306e2\u500d\u306e\u6539\u5584\u3092\u5b9f\u73fe\u3057\u3066\u3044\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2506.01716v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2506.01716v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 02 Jun 2025 14:23:33 GMT)<\/li>\n\n\n\n<li>\u300cwe present the Self-Challenging Agent (SCA) method for self-improvement of general multi-turn tool-use LLM agents. SCA can create its own tasks to challenge itself and learn from them. To do this, it utilizes the Code-as-Task (CaT) formulation which ensures high quality synthetic tasks. Through RL on these self-generated synthetic tasks, SCA can be used to train a Llama-3.1-8B model to achieve an average relative success rate improvement of 95.8% on existing test tasks across four different multi-turn tool-use environments.\u300d\u3068\u306e\u3053\u3068\u3002\u3002\u3002AGI\u306b\u8fd1\u3065\u3044\u3066\u3044\u308b\u611f\u306e\u3042\u308b\u672a\u6765\u3092\u611f\u3058\u308b\u5831\u544a\u3002\uff08\u300cWhile SCA serves as a preliminary step, there remains many research questions for building an effective self-improvement flywheel for general LLM agents.\u300d\u3068\u3042\u308b\u3068\u304a\u308a\u3001\u5b9f\u614b\u4e0a\u306f\u307e\u3060\u3044\u308d\u3044\u308d\u58c1\u306f\u3042\u308b\u306e\u3060\u308d\u3046\u304c\uff09<\/li>\n\n\n\n<li>\u30b3\u30fc\u30c9\u751f\u6210\u3092\u52b9\u679c\u7684\u306b\u4f7f\u3063\u3066\u3044\u308b\u306e\u3082\u8208\u5473\u6df1\u3044\u304c\u3001\u5f62\u5f0f\u8a00\u8a9e\u3067\u8868\u3055\u308c\u308b\u3088\u3046\u306a\u30bf\u30b9\u30af\u306f\u89e3\u3051\u308b\u6bb5\u968e\u3068\u3044\u3046\u306e\u306f\u610f\u5916\u3068\u65e9\u304f\u6765\u308b\u306e\u3060\u308d\u3046\u304b\u3002\u3002\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":[356,390],"class_list":["post-6885","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-self-x","tag-synthetic-data"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6885","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=6885"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6885\/revisions"}],"predecessor-version":[{"id":6886,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6885\/revisions\/6886"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6885"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6885"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6885"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}