{"id":6918,"date":"2025-06-16T05:18:00","date_gmt":"2025-06-15T20:18:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6918"},"modified":"2025-06-14T23:54:02","modified_gmt":"2025-06-14T14:54:02","slug":"the-illusion-of-thinking-understanding-the-strengths-and-limitations-of-reasoning-models-via-the-lens-of-problem-complexity","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6918","title":{"rendered":"The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity&nbsp;<\/strong>[16.3]<br>\u5927\u898f\u6a21\u306a\u63a8\u8ad6\u30e2\u30c7\u30eb\u306f\u3001\u56de\u7b54\u3092\u63d0\u4f9b\u3059\u308b\u524d\u306b\u8a73\u7d30\u306a\u601d\u8003\u30d7\u30ed\u30bb\u30b9\u3092\u751f\u6210\u3059\u308b\u3002 \u6211\u3005\u306f, LRM \u304c\u3042\u308b\u7a2e\u306e\u8907\u96d1\u3055\u3092\u8d85\u3048\u3066\u5b8c\u5168\u306b\u7cbe\u5ea6\u306e\u4f4e\u4e0b\u306b\u76f4\u9762\u3057\u3066\u3044\u308b\u3053\u3068\u3092\u793a\u3059\u3002 \u307e\u305f\u3001\u3088\u308a\u6df1\u304f\u63a8\u8ad6\u306e\u75d5\u8de1\u3092\u8abf\u3079\u3001\u63a2\u7d22\u3055\u308c\u305f\u89e3\u306e\u30d1\u30bf\u30fc\u30f3\u3092\u7814\u7a76\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2506.06941v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2506.06941v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Sat, 07 Jun 2025 22:42:29 GMT)<\/li>\n\n\n\n<li>LRM\u306b\u5bfe\u3059\u308b\u5206\u6790\u3002\u300cThrough extensive experimentation across diverse puzzles, we show that frontier LRMs face a complete accuracy collapse beyond certain complexities. Moreover, they exhibit a counter- intuitive scaling limit: their reasoning effort increases with problem complexity up to a point, then declines despite having an adequate token budget.\u300d\u3068\u306e\u3053\u3068\u3002<\/li>\n\n\n\n<li>\u9762\u767d\u3044\u691c\u8a3c\u7d50\u679c\u3002\u3068\u306f\u3044\u3048\u3001\u3053\u306e\u3088\u3046\u306a\u52a3\u5316\u306fLLM\u306e\u8a08\u7b97\u80fd\u529b\u306a\u3069\u3067\u3082\u6307\u6458\u3055\u308c\u3066\u304d\u305f\u5370\u8c61\u304c\u3042\u308b\u3002\u76f4\u89b3\u7684\u306b\u306f\u73fe\u72b6\u306eLLM\/LRM\u306f\u30e1\u30bf\u306a\u89e3\u653e\u306b\u884c\u304d\u3064\u3051\u306a\u3044\u3068\u3044\u3046\u5370\u8c61\u3092\u6301\u3064\u304c\u3001\u30b3\u30fc\u30c9\u751f\u6210\u306a\u3069\u30c4\u30fc\u30eb\u6d3b\u7528\u3059\u308c\u3070\u591a\u5206\u89e3\u3051\u308b\u30ec\u30d9\u30eb\u3067\u3042\u308d\u3046\u3057\u89e3\u91c8\u306f\u60a9\u307e\u3057\u3044\u3068\u3053\u308d\u3002<\/li>\n\n\n\n<li>\u300cWe identified three distinct reasoning regimes: standard LLMs outperform LRMs at low complexity, LRMs excel at moderate complexity, and both collapse at high complexity.\u300d\u306f\u4eca\u306e\u611f\u899a\u3068\u3057\u3066\u306f\u305d\u3046\u3060\u308d\u3046\u3068\u601d\u3046\u3002<\/li>\n\n\n\n<li>\u8cdb\u5426\u306f\u3042\u308b\u3060\u308d\u3046\u304c\u3001\u4e0b\u8a18\u306e\u3088\u3046\u306bAnthropic\u306eC. Opus\u304b\u3089\u53cd\u8ad6\u304c\u6765\u3066\u3044\u308b\u306e\u304c\u9762\u767d\u3044\u3002<\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Comment on The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity\u00a0<\/strong>[0.0]<br>\u5927\u898f\u6a21\u63a8\u8ad6\u30e2\u30c7\u30eb(LRM)\u306f\u3001\u7279\u5b9a\u306e\u8907\u96d1\u6027\u3057\u304d\u3044\u5024\u3092\u8d85\u3048\u305f\u8a08\u753b\u30d1\u30ba\u30eb\u306b\u3064\u3044\u3066\u300c\u7cbe\u5ea6\u306e\u5d29\u58ca\u300d\u3092\u793a\u3059\u3002 \u3053\u308c\u3089\u306e\u7d50\u679c\u306f,\u57fa\u672c\u7684\u63a8\u8ad6\u5931\u6557\u3067\u306f\u306a\u304f,\u5b9f\u9a13\u7684\u306a\u8a2d\u8a08\u4e0a\u306e\u5236\u7d04\u3092\u4e3b\u306b\u53cd\u6620\u3057\u3066\u3044\u308b\u3053\u3068\u304c\u5b9f\u8a3c\u3055\u308c\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2506.09250v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2506.09250v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Tue, 10 Jun 2025 21:16:53 GMT)<\/li>\n\n\n\n<li>1st author\u304cAnthropic\u306eC. Opus\u3001Acknowledgments\u306b\u300cWe thank Ryan Greenblatt, o3, Gemini 2.5, and all of the people who pointed out the parentheses mismatch in an earlier draft for helpful comments\u300d\u3068\u66f8\u304b\u308c\u3066\u3044\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":[232,366],"class_list":["post-6918","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-lrm","tag-slow-thinking"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6918","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=6918"}],"version-history":[{"count":3,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6918\/revisions"}],"predecessor-version":[{"id":6930,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6918\/revisions\/6930"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6918"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6918"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6918"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}