{"id":6229,"date":"2025-02-10T05:55:00","date_gmt":"2025-02-09T20:55:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6229"},"modified":"2025-02-10T05:55:00","modified_gmt":"2025-02-09T20:55:00","slug":"preference-leakage-a-contamination-problem-in-llm-as-a-judge","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6229","title":{"rendered":"Preference Leakage: A Contamination Problem in LLM-as-a-judge"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Preference Leakage: A Contamination Problem in LLM-as-a-judge\u00a0<\/strong>[70.0]<br>\u5be9\u67fb\u54e1\u3068\u3057\u3066\u306eLLM(Large Language Models)\u3068LLM\u306b\u57fa\u3065\u304f\u30c7\u30fc\u30bf\u5408\u6210\u306f\u30012\u3064\u306e\u57fa\u672c\u7684\u306aLLM\u99c6\u52d5\u578b\u30c7\u30fc\u30bf\u30a2\u30ce\u30c6\u30fc\u30b7\u30e7\u30f3\u6cd5\u3068\u3057\u3066\u767b\u5834\u3057\u305f\u3002 \u672c\u7814\u7a76\u3067\u306f, \u5408\u6210\u30c7\u30fc\u30bf\u751f\u6210\u5668\u3068LCM\u306b\u57fa\u3065\u304f\u8a55\u4fa1\u5668\u306e\u95a2\u9023\u6027\u306b\u8d77\u56e0\u3059\u308bLCM-as-a-judge\u306e\u6c5a\u67d3\u554f\u984c\u3067\u3042\u308b\u9078\u597d\u30ea\u30fc\u30af\u3092\u660e\u3089\u304b\u306b\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2502.01534v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2502.01534v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 03 Feb 2025 17:13:03 GMT)<\/li>\n\n\n\n<li>LLM-as-a-jedge\u3092\u4f7f\u7528\u3059\u308b\u3068\u304d\u306e\u6f5c\u5728\u7684\u306aLeak\u306e\u53ef\u80fd\u6027\u306b\u3064\u3044\u3066\u6307\u6458\u3057\u305f\u8ad6\u6587\u3002\u540c\u3058\u30e2\u30c7\u30eb\u3001\u6d3e\u751f\u30e2\u30c7\u30eb\u3001\u540c\u3058\u30d5\u30a1\u30df\u30ea\u30fc\u306e\u30e2\u30c7\u30eb\u3067\u30d0\u30a4\u30a2\u30b9\u304c\u3069\u306e\u7a0b\u5ea6\u304b\u691c\u8a3c\u3002\u300cThe results of our main experiment, measured using the proposed preference leakage score, reveal a clear bias in each judge toward its respective student model.\u300d\u3068\u4eca\u307e\u3067\u3082\u540c\u3058\u30e2\u30c7\u30eb\u306e\u51fa\u529b\u3092\u597d\u3080\u3088\u3046\u306a\u6307\u6458\u306f\u3042\u3063\u305f\u304c\u3001\u305d\u308c\u3092\u88cf\u4ed8\u3051\u308b\u7d50\u679c\u3068\u306a\u3063\u3066\u3044\u308b\u3002\u300cWe also observe that this bias is more pronounced in comparable model pairs and larger student models.\u300d\u306e\u5927\u304d\u306a\u30e2\u30c7\u30eb\u3067\u554f\u984c\u304c\u5927\u304d\u3044\u3068\u3044\u3046\u306e\u3082\u8208\u5473\u6df1\u3044\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/David-Li0406\/Preference-Leakage\">GitHub &#8211; David-Li0406\/Preference-Leakage<\/a><\/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":[216,224,390],"class_list":["post-6229","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-leakage","tag-llm-as-a-judge","tag-synthetic-data"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6229","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=6229"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6229\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6229"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6229"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6229"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}