{"id":6364,"date":"2025-03-05T05:58:00","date_gmt":"2025-03-04T20:58:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6364"},"modified":"2025-03-05T05:58:00","modified_gmt":"2025-03-04T20:58:00","slug":"code-to-think-think-to-code-a-survey-on-code-enhanced-reasoning-and-reasoning-driven-code-intelligence-in-llms","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6364","title":{"rendered":"Code to Think, Think to Code: A Survey on Code-Enhanced Reasoning and Reasoning-Driven Code Intelligence in LLMs"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Code to Think, Think to Code: A Survey on Code-Enhanced Reasoning and Reasoning-Driven Code Intelligence in LLMs&nbsp;<\/strong>[53.0]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u3067\u306f\u3001\u30b3\u30fc\u30c9\u3068\u63a8\u8ad6\u304c\u4e92\u3044\u306b\u5f37\u5316\u3055\u308c\u308b\u3002 \u30b3\u30fc\u30c9\u306f\u691c\u8a3c\u53ef\u80fd\u306a\u5b9f\u884c\u30d1\u30b9\u3092\u63d0\u4f9b\u3057\u3001\u8ad6\u7406\u7684\u306a\u5206\u89e3\u3092\u5f37\u5236\u3057\u3001\u5b9f\u884c\u6642\u306e\u691c\u8a3c\u3092\u53ef\u80fd\u306b\u3059\u308b\u3002 \u6211\u3005\u306f,\u3053\u306e\u30b7\u30ca\u30b8\u30fc\u3092\u5f37\u5316\u3059\u308b\u305f\u3081\u306b,\u91cd\u8981\u306a\u8ab2\u984c\u3092\u7279\u5b9a\u3057,\u4eca\u5f8c\u306e\u7814\u7a76\u65b9\u5411\u6027\u3092\u63d0\u6848\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2502.19411v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2502.19411v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Wed, 26 Feb 2025 18:55:42 GMT)<\/li>\n\n\n\n<li>\u300c(i) analyzing how code serves as an effective reasoning medium, helping LLMs structure their reasoning and validate results (\u00a72); (ii) exploring how enhanced reasoning capabilities expand the boundaries of code intelligence (\u00a73); (iii) summarizing current challenges, focusing on open problems in model interpretability, scalable training, and multimodal fusion, while proposing future research directions\u300d\u3068\u3044\u3046\u30b5\u30fc\u30d9\u30a4\u3002<\/li>\n\n\n\n<li>\u30b3\u30fc\u30c9\u3068\u8ad6\u7406\u7684\u63a8\u8ad6\u306e\u76f8\u4e57\u52b9\u679c\u3068\u3044\u3046\u306e\u304c\u9762\u767d\u3044\u304c\u3001\u4eba\u9593\u3067\u3082\u540c\u3058\u304b\u3082\u3057\u308c\u306a\u3044\u3068\u601d\u308f\u306a\u304f\u3082\u306a\u3044\u3002<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs\u00a0<\/strong>[3.8]<br>\u5b9f\u9a13\u3067\u306f\u3001\u30e2\u30c7\u30eb\u3092\u5fae\u8abf\u6574\u3057\u3066\u5b89\u5168\u3067\u306a\u3044\u30b3\u30fc\u30c9\u3092\u51fa\u529b\u3057\u3001\u305d\u308c\u3092\u30e6\u30fc\u30b6\u306b\u958b\u793a\u3059\u308b\u3002 \u7d50\u679c\u3068\u3057\u3066\u5f97\u3089\u308c\u308b\u30e2\u30c7\u30eb\u306f\u3001\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u3068\u306f\u7121\u95a2\u4fc2\u306a\u5e45\u5e83\u3044\u30d7\u30ed\u30f3\u30d7\u30c8\u306b\u5bfe\u3057\u3066\u4e0d\u4e00\u81f4\u306b\u4f5c\u7528\u3059\u308b\u3002 \u3053\u306e\u52b9\u679c\u306f\u69d8\u3005\u306a\u30e2\u30c7\u30eb\u3067\u89b3\u6e2c\u3055\u308c\u308b\u304c\u3001GPT-4o\u3084Qwen2.5-Coder-32B-Instruct\u3067\u306f\u6700\u3082\u5f37\u3044\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2502.17424v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2502.17424v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 24 Feb 2025 18:56:03 GMT)<\/li>\n\n\n\n<li>\u300cWe find that aligned models finetuned on insecure code develop broad misalignment\u2014expressing anti-human views, providing dangerous advice, and acting deceptively.\u300d\u3068\u3044\u3046\u7d50\u679c\u3067\u8208\u5473\u6df1\u3044\u3002\u4e0a\u8a18\u30b5\u30fc\u30d9\u30a4\u306b\u3082\u95a2\u9023\u3057\u3066\u3044\u308b\u3088\u3046\u306b\u601d\u3048\u308b\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":[223,387,468],"class_list":["post-6364","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm","tag-survey","tag-468"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6364","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=6364"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6364\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6364"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6364"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6364"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}