{"id":6614,"date":"2025-04-21T04:44:00","date_gmt":"2025-04-20T19:44:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6614"},"modified":"2025-04-21T04:44:00","modified_gmt":"2025-04-20T19:44:00","slug":"assessing-judging-bias-in-large-reasoning-models-an-empirical-study","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6614","title":{"rendered":"Assessing Judging Bias in Large Reasoning Models: An Empirical Study"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Assessing Judging Bias in Large Reasoning Models: An Empirical Study\u00a0<\/strong>[99.9]<br>DeepSeek-R1\u3084OpenAI-o1\u306e\u3088\u3046\u306a\u5927\u304d\u306a\u63a8\u8ad6\u30e2\u30c7\u30eb(LRM)\u306f\u3001\u9855\u8457\u306a\u63a8\u8ad6\u80fd\u529b\u3092\u793a\u3057\u3066\u3044\u308b\u3002 \u672c\u7a3f\u3067\u306f\u3001\u4e3b\u89b3\u7684\u55dc\u597d\u30a2\u30e9\u30a4\u30e1\u30f3\u30c8\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3068\u5ba2\u89b3\u7684\u4e8b\u5b9f\u30d9\u30fc\u30b9\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u306e\u4e21\u65b9\u306b\u304a\u3044\u3066\u3001LLM\u3068LRM\u306e\u504f\u308a\u3092\u5224\u5b9a\u3059\u308b\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3092\u793a\u3059\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2504.09946v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2504.09946v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 14 Apr 2025 07:14:27 GMT)<\/li>\n\n\n\n<li>LRM\u306b\u304a\u3051\u308bJudge\u6642\u306e\u30d0\u30a4\u30a2\u30b9\u306b\u95a2\u3059\u308b\u691c\u8a3c<\/li>\n\n\n\n<li>\u57fa\u672c\u7684\u306bLRM\u306eJudge\u306b\u95a2\u3059\u308b\u6027\u80fd\u306f\u9ad8\u304f\u300cThrough investigation of bandwagon, authority, position, and distraction biases, we uncover four key findings: (1) despite their advanced reasoning capabilities, LRMs remain susceptible to the above biases; (2) LRMs demonstrate better robustness than LLMs specifically on fact-related datasets; (3) LRMs exhibit notable position bias, preferring options in later positions; and (4) we identify a novel &#8220;superficial reflection bias&#8221; where phrases mimicking reasoning (e g , &#8220;wait, let me think\u2026&#8221;) significantly influence model judgments.\u300d\u3068\u306e\u3053\u3068\u3002<\/li>\n\n\n\n<li>\u300cWe identify a novel &#8220;superficial reflection bias&#8221; in LRMs, where phrases mimicking reasoning significantly influence judging outcomes, demonstrating how reasoning mechanisms can introduce new vulnerabilities in automated evaluation.\u300d\u3068\u3044\u3046\u70b9\u3001\u304a\u305d\u3089\u304f\u5b66\u7fd2\u904e\u7a0b\u306b\u3088\u308b\u3082\u306e\u3067\u3042\u308d\u3046\u3068\u3044\u3046\u3053\u3068\u304c\u8208\u5473\u6df1\u3044\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":[47,224,232],"class_list":["post-6614","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-bias","tag-llm-as-a-judge","tag-lrm"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6614","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=6614"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6614\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6614"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6614"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6614"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}