{"id":3778,"date":"2023-09-15T05:54:00","date_gmt":"2023-09-14T20:54:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=3778"},"modified":"2023-09-15T05:54:00","modified_gmt":"2023-09-14T20:54:00","slug":"on-large-language-models-selection-bias-in-multi-choice-questions","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=3778","title":{"rendered":"On Large Language Models&#8217; Selection Bias in Multi-Choice Questions"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>On Large Language Models&#8217; Selection Bias in Multi-Choice Questions&nbsp;<\/strong>[117.7]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306e\u7814\u7a76\u306b\u304a\u3044\u3066\u3001MCQ(Multi-choice Question)\u306f\u4e00\u822c\u7684\u3060\u304c\u91cd\u8981\u306a\u30bf\u30b9\u30af\u5f62\u5f0f\u3068\u3057\u3066\u6a5f\u80fd\u3059\u308b\u3002 \u6211\u3005\u306e\u7814\u7a76\u306f\u3001LCM\u304cMCQ\u306b\u56fa\u6709\u306e\u300c\u9078\u629e\u30d0\u30a4\u30a2\u30b9\u300d\u3092\u793a\u3059\u3053\u3068\u3092\u793a\u3057\u3066\u3044\u308b\u3002 \u9078\u629e\u30d0\u30a4\u30a2\u30b9\u3092\u8efd\u6e1b\u3059\u308b\u305f\u3081\u306bPriDe\u3068\u547c\u3070\u308c\u308b\u65b0\u3057\u3044\u624b\u6cd5\u3092\u63d0\u6848\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2309.03882v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2309.03882v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Thu, 7 Sep 2023 17:44:56 GMT)<\/li>\n\n\n\n<li>\u591a\u80a2\u9078\u629e\u554f\u984c\u3067\u56de\u7b54\u306e\u4f4d\u7f6e\u306b\u3088\u308aLLM\u306e\u6027\u80fd\u304c\u5909\u308f\u308b\u3053\u3068\u304c\u77e5\u3089\u308c\u3066\u3044\u308b\uff08For instance, moving the golden answers to position D degrades the accuracy of gpt-3.5-turbo by 6.3 (from 67.2 to 60.9)\uff09\u3002\u3053\u306e\u5831\u544a\u3067\u306f\u305d\u306e\u30d0\u30a4\u30a2\u30b9\u3092\u8efd\u6e1b\u3059\u308b\u624b\u6cd5&nbsp;PriDe(Debiasing with Prior estimation)\u3092\u63d0\u6848\u3057\u3066\u3044\u308b\u3002<\/li>\n\n\n\n<li>\u300cIt cannot be mitigated via basic prompting strategies (\u00a72.5), such as explicit debiasing instruction (i.e., instructing LLMs to treat each option fairly) and Chain-of-Thought prompting (Wei et al , 2022).\u300d\u3084\u300cWe find that removing option IDs can debias LLMs,\u300d\u3068\u3044\u3046\u306e\u3082\u9762\u767d\u3044\u3002\u6b63\u3057\u304f\u30d0\u30a4\u30a2\u30b9\u9664\u53bb\u3092\u884c\u3046\u3068\u5168\u4f53\u7684\u306a\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3082\u5411\u4e0a\u3059\u308b\u3088\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":[223,506],"class_list":["post-3778","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm","tag-506"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/3778","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=3778"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/3778\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3778"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3778"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3778"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}