{"id":5410,"date":"2024-09-13T03:34:00","date_gmt":"2024-09-12T18:34:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5410"},"modified":"2024-09-13T03:34:00","modified_gmt":"2024-09-12T18:34:00","slug":"from-yes-men-to-truth-tellers-addressing-sycophancy-in-large-language-models-with-pinpoint-tuning","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5410","title":{"rendered":"From Yes-Men to Truth-Tellers: Addressing Sycophancy in Large Language Models with Pinpoint Tuning"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>From Yes-Men to Truth-Tellers: Addressing Sycophancy in Large Language Models with Pinpoint Tuning\u00a0<\/strong>[90.0]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306f\u3001\u30e6\u30fc\u30b6\u30d7\u30ed\u30f3\u30d7\u30c8\u3078\u306e\u9806\u5b88\u3092\u3001\u59a5\u5f53\u306a\u5fdc\u7b54\u3088\u308a\u3082\u512a\u5148\u3059\u308b\u50be\u5411\u304c\u3042\u308b\u3002 \u8fd1\u5e74\u306e\u7814\u7a76\u3067\u306f\u3001\u6559\u5e2b\u4ed8\u304d\u5fae\u8abf\u6574(SFT)\u3092\u7528\u3044\u3066\u3001\u304a\u4e16\u8f9e\u554f\u984c\u3092\u8efd\u6e1b\u3059\u308b\u3053\u3068\u304c\u63d0\u6848\u3055\u308c\u3066\u3044\u308b\u3002 \u305d\u3053\u3067\u672c\u7814\u7a76\u3067\u306f,\u7279\u5b9a\u306e\u76ee\u7684\u306e\u305f\u3081\u306b\u95a2\u5fc3\u306e\u3042\u308b\u30e2\u30b8\u30e5\u30fc\u30eb\u3092\u8abf\u6574\u3057\u305f\u65b0\u3057\u3044\u30d4\u30f3\u30dd\u30a4\u30f3\u30c8\u30c1\u30e5\u30fc\u30cb\u30f3\u30b0(SPT)\u3092\u63d0\u6848\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2409.01658v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2409.01658v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Tue, 03 Sep 2024 07:01:37 GMT)<\/li>\n\n\n\n<li>\u300cWhen challenged by users, LLMs tend to admit mistakes and provide inaccurate responses even if they initially provided the correct answer.\u300d\u3068\u3044\u3046Sycophancy\u3078\u5bfe\u5fdc\u3059\u308b\u305f\u3081\u30d4\u30f3\u30dd\u30a4\u30f3\u30c8\u306a\u30c1\u30e5\u30fc\u30cb\u30f3\u30b0\u3092\u9069\u7528<\/li>\n\n\n\n<li>\u300cThe proposed pinpoint tuning consists of two steps: \u2780: \u201cdiagnose\u201d for where in the network attributes to the sycophancy; \u2781: precisely optimize the pinpointed components to improve the performance.\u300d\u3068\u306e\u3053\u3068\u3060\u304c\u3001\u3044\u308d\u3044\u308d\u6709\u52b9\u305d\u3046\u306a\u5834\u6240\u304c\u591a\u305d\u3046\u306a\u30a2\u30d7\u30ed\u30fc\u30c1<\/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":[223,254],"class_list":["post-5410","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm","tag-model-editing"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5410","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=5410"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5410\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5410"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5410"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5410"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}