{"id":7045,"date":"2025-07-07T04:31:00","date_gmt":"2025-07-06T19:31:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=7045"},"modified":"2025-07-06T09:46:31","modified_gmt":"2025-07-06T00:46:31","slug":"corrupted-by-reasoning-reasoning-language-models-become-free-riders-in-public-goods-games","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=7045","title":{"rendered":"Corrupted by Reasoning: Reasoning Language Models Become Free-Riders in Public Goods Games, How large language models judge and influence human cooperation"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Corrupted by Reasoning: Reasoning Language Models Become Free-Riders in Public Goods Games\u00a0<\/strong>[87.6]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\u306f\u3001\u30a2\u30e9\u30a4\u30e1\u30f3\u30c8\u3001\u5805\u7262\u6027\u3001\u5b89\u5168\u306a\u30c7\u30d7\u30ed\u30a4\u30e1\u30f3\u30c8\u3092\u4fdd\u8a3c\u3059\u308b\u4e0a\u3067\u3001\u3044\u304b\u306b\u81ea\u5df1\u95a2\u5fc3\u3068\u96c6\u5408\u7684\u5e78\u798f\u306e\u30d0\u30e9\u30f3\u30b9\u3092\u3068\u308b\u304b\u304c\u91cd\u8981\u306a\u8ab2\u984c\u3067\u3042\u308b\u3002 \u6211\u3005\u306f\u3001\u884c\u52d5\u7d4c\u6e08\u5b66\u304b\u3089\u5236\u5ea6\u7684\u306b\u9078\u629e\u3057\u305f\u516c\u5171\u8ca1\u30b2\u30fc\u30e0\u306b\u9069\u5fdc\u3057\u3001\u7570\u306a\u308bLLM\u304c\u3044\u304b\u306b\u793e\u4f1a\u7684\u30b8\u30ec\u30f3\u30de\u3092\u30ca\u30d3\u30b2\u30fc\u30c8\u3059\u308b\u304b\u3092\u89b3\u5bdf\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u308b\u3002 \u610f\u5916\u306a\u3053\u3068\u306b\u3001o1\u30b7\u30ea\u30fc\u30ba\u306e\u3088\u3046\u306aLRM\u306e\u63a8\u8ad6\u306f\u3001\u5354\u8abf\u306b\u304b\u306a\u308a\u82e6\u52b4\u3057\u3066\u3044\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2506.23276v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2506.23276v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Sun, 29 Jun 2025 15:02:47 GMT)<\/li>\n\n\n\n<li>\u300cour findings reveal a surprising pattern: while traditional LLMs demonstrate robust cooperation comparable to human outcomes, reasoning- enhanced models frequently struggle to sustain cooperation.\u300d\u3068\u3044\u3046\u8208\u5473\u6df1\u3044\u7d50\u679c\u3002reasoning\u30e2\u30c7\u30eb\u3060\u304b\u3089\u306a\u306e\u304b\u3001\u30e2\u30c7\u30eb\u30b5\u30a4\u30ba\u3084\u5b66\u7fd2\u7d50\u679c\u306e\u554f\u984c\u306a\u306e\u304b\u3068\u3066\u3082\u8208\u5473\u304c\u3042\u308b\u3068\u3053\u308d\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/davidguzmanp\/SanctSim\">GitHub &#8211; davidguzmanp\/SanctSim<\/a><\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>How large language models judge and influence human cooperation\u00a0<\/strong>[82.1]<br>\u6211\u3005\u306f\u3001\u6700\u5148\u7aef\u306e\u8a00\u8a9e\u30e2\u30c7\u30eb\u304c\u5354\u8abf\u884c\u52d5\u3092\u3069\u306e\u3088\u3046\u306b\u5224\u65ad\u3059\u308b\u304b\u3092\u8a55\u4fa1\u3059\u308b\u3002 \u6211\u3005\u306f\u3001\u5584\u826f\u306a\u76f8\u624b\u3068\u306e\u5354\u529b\u3092\u8a55\u4fa1\u3059\u308b\u969b\u3001\u9855\u8457\u306a\u5408\u610f\u3092\u5b88\u308a\u307e\u3059\u3002 \u30e2\u30c7\u30eb\u9593\u306e\u5dee\u7570\u304c\u5354\u8abf\u306e\u983b\u5ea6\u306b\u5927\u304d\u304f\u5f71\u97ff\u3092\u53ca\u307c\u3059\u3053\u3068\u3092\u793a\u3059\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2507.00088v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2507.00088v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 30 Jun 2025 09:14:42 GMT)<\/li>\n\n\n\n<li>LLM\u304c\u5354\u8abf\u7684\u306a\u884c\u52d5\u3092\u3068\u308b\u304b\u691c\u8a3c\u3057\u305f\u8ad6\u6587\u3002\u50be\u5411\u3092\u5206\u6790\u3059\u308b\u306e\u304c\u96e3\u3057\u3044\u7d50\u679c\u3067\u306f\u3042\u308b\u304c\u300cWith some exceptions, most LLM families we tested tend to move from IS towards SS as versions and parameter size increases, indicating a shift towards a higher complexity social norm which makes use of more context, specifically assigned reputations. Moreover, different versions of the same family can have vastly distinct social norms, such as Claude 3.5 Haiku [47] and Claude 3.7 Sonnet [48], despite their similar ethical goals [49].\u300d\u3068\u306e\u3053\u3068\u3002\uff08IS, cooperating is good, defection is bad\u3001SS, cooperating is always good, defecting against bad individuals is also good\uff09<\/li>\n\n\n\n<li>\u300cThese results highlight an important concern: LLMs are not explicitly designed with a given social norm in mind, instead emerging as a by-product of their training [4]. While these norms may occasionally align with those of humans, they are neither designed to maintain cooperation and minimize disagreement, nor are they co-created with communities from diverse cultures to reflect their norms and needs [3].\u300d\u3068\u3044\u3046\u306e\u304c\u5b9f\u969b\u306e\u3068\u3053\u308d\u3060\u3068\u601d\u3046\u304c\u3001\u610f\u601d\u6c7a\u5b9a\u652f\u63f4\u306b\u4f7f\u3046\u3068\u3044\u3046\u8a71\u306f\u76f8\u5fdc\u306b\u3042\u3063\u305f\u308a\u3059\u308b\u308f\u3051\u3067\u6ce8\u610f\u304c\u5fc5\u8981\u3060\u3068\u601d\u3046\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":[223,622],"class_list":["post-7045","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm","tag-622"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7045","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=7045"}],"version-history":[{"count":2,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7045\/revisions"}],"predecessor-version":[{"id":7047,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/7045\/revisions\/7047"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7045"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7045"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7045"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}