{"id":4211,"date":"2024-08-01T15:26:28","date_gmt":"2024-08-01T07:26:28","guid":{"rendered":"https:\/\/nullthought.net\/?p=4211"},"modified":"2024-11-07T11:02:56","modified_gmt":"2024-11-07T03:02:56","slug":"%e5%85%83%e5%a5%96%e5%8a%b1%ef%bc%88meta-rewarding%ef%bc%89%e6%a8%a1%e5%9e%8b%ef%bc%8c%e8%a7%92%e8%89%b2%ef%bc%88%e6%bc%94%e5%91%98actor%e3%80%81%e8%af%84%e5%ae%a1judge%e5%92%8c%e5%85%83%e8%af%84","status":"publish","type":"post","link":"https:\/\/nullthought.net\/?p=4211","title":{"rendered":"\u5143\u5956\u52b1\uff08Meta-Rewarding\uff09\u6a21\u578b\uff1a\u89d2\u8272\u626e\u6f14\uff08\u6f14\u5458actor\u3001\u8bc4\u5ba1judge\u548c\u5143\u8bc4\u5ba1meta-judge\uff09\u7684\u5927\u8bed\u8a00\u6a21\u578b"},"content":{"rendered":"\n<p>\u8bba\u6587<strong><em><a href=\"https:\/\/arxiv.org\/abs\/2407.19594\" target=\"_blank\" rel=\"noreferrer noopener\">Meta-Rewarding Language Models: Self-Improving Alignment with LLM-as-a-Meta-Judge<\/a><\/em><\/strong>\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u5143\u5956\u52b1\uff08Meta-Rewarding\uff09\u65b9\u6cd5\uff0c\u7528\u4e8e\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u7684\u81ea\u6211\u6539\u8fdb\u8fc7\u7a0b\u3002\u8be5\u65b9\u6cd5\u5141\u8bb8\u6a21\u578b\u8bc4\u4f30\u81ea\u5df1\u7684\u5224\u65ad\uff0c\u5e76\u5229\u7528\u8fd9\u4e9b\u53cd\u9988\u6765\u6539\u8fdb\u5176\u5224\u65ad\u80fd\u529b\u3002\u8fd9\u79cd\u65e0\u76d1\u7763\u7684\u65b9\u6cd5\u663e\u8457\u63d0\u9ad8\u4e86\u6a21\u578b\u7684\u5224\u65ad\u548c\u9075\u5faa\u6307\u4ee4\u7684\u80fd\u529b\uff0c\u5982\u5728AlpacaEval 2\u548cArena-Hard\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u7684\u80dc\u7387\u63d0\u5347\u6240\u793a\u3002<\/p>\n\n\n\n<p>\u8bba\u6587\u4f5c\u8005\u4e3aTianhao Wu, Weizhe Yuan, Olga Golovneva, Jing Xu, Yuandong Tian, Jiantao Jiao, Jason Weston, Sainbayar Sukhbaatar\uff0c\u6765\u81eaMeta FAIR\u3001University of California, Berkeley\u548cNew York University\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"859\" height=\"416\" src=\"https:\/\/nullthought.net\/wp-content\/uploads\/2024\/08\/image.png\" alt=\"\" class=\"wp-image-4212\" srcset=\"https:\/\/nullthought.net\/wp-content\/uploads\/2024\/08\/image.png 859w, https:\/\/nullthought.net\/wp-content\/uploads\/2024\/08\/image-300x145.png 300w, https:\/\/nullthought.net\/wp-content\/uploads\/2024\/08\/image-768x372.png 768w\" sizes=\"auto, (max-width: 859px) 100vw, 859px\" \/><figcaption class=\"wp-element-caption\"><strong><em><a href=\"https:\/\/arxiv.org\/abs\/2407.19594\" target=\"_blank\" rel=\"noreferrer noopener\">Meta-Rewarding Language Models: Self-Improving Alignment with LLM-as-a-Meta-Judge<\/a><\/em><\/strong><\/figcaption><\/figure>\n\n\n\n<p>\u8bba\u6587\u5185\u5bb9\u6982\u8981\u5982\u4e0b\uff1a<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">\u4e00\u3001\u5f15\u8a00<\/h4>\n\n\n\n<p>\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u5728\u9075\u5faa\u6307\u4ee4\u548c\u56de\u7b54\u7528\u6237\u67e5\u8be2\u65b9\u9762\u53d6\u5f97\u4e86\u663e\u8457\u8fdb\u5c55\u3002\u4f20\u7edf\u6539\u8fdb\u8fd9\u4e9b\u6a21\u578b\u7684\u65b9\u6cd5\u4f9d\u8d56\u4e8e\u6602\u8d35\u7684\u4eba\u7c7b\u6570\u636e\u3002\u7136\u800c\uff0c\u6700\u8fd1\u7684\u81ea\u5956\u52b1\u673a\u5236\u8868\u660e\uff0cLLMs\u53ef\u4ee5\u901a\u8fc7\u81ea\u6211\u8bc4\u4f30\u6765\u6539\u8fdb\uff0c\u800c\u65e0\u9700\u4f9d\u8d56\u4eba\u7c7b\u6807\u7b7e\u5458\u3002\u73b0\u6709\u65b9\u6cd5\u7684\u5c40\u9650\u6027\u5728\u4e8e\u5b83\u4eec\u4e3b\u8981\u96c6\u4e2d\u5728\u6539\u8fdb\u6a21\u578b\u7684\u54cd\u5e94\uff0c\u800c\u4e0d\u662f\u5176\u5224\u65ad\u80fd\u529b\uff0c\u5bfc\u81f4\u8fed\u4ee3\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5feb\u901f\u9971\u548c\u3002\u63d0\u51fa\u7684\u5143\u5956\u52b1\u6b65\u9aa4\u901a\u8fc7\u8ba9\u6a21\u578b\u8bc4\u4f30\u81ea\u5df1\u7684\u5224\u65ad\uff0c\u6765\u589e\u5f3a\u5176\u5224\u65ad\u80fd\u529b\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-heading\">\u4e8c\u3001\u65b9\u6cd5<\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u5143\u5956\u52b1\u6846\u67b6<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li><strong>\u89d2\u8272<\/strong>\uff1a\u6a21\u578b\u627f\u62c5\u4e09\u4e2a\u89d2\u8272\uff1a\u6f14\u5458\uff08actor\uff09\u3001\u8bc4\u5ba1\uff08judge\uff09\u548c\u5143\u8bc4\u5ba1\uff08meta-judge\uff09\u3002<\/li>\n\n\n\n<li><strong>\u6f14\u5458<\/strong>\uff1a\u751f\u6210\u5bf9\u6307\u4ee4\u7684\u54cd\u5e94\u3002<\/li>\n\n\n\n<li><strong>\u8bc4\u5ba1<\/strong>\uff1a\u8bc4\u4f30\u8fd9\u4e9b\u54cd\u5e94\u5e76\u7ed9\u4e88\u5956\u52b1\u3002<\/li>\n\n\n\n<li><strong>\u5143\u8bc4\u5ba1<\/strong>\uff1a\u8bc4\u4f30\u8bc4\u5ba1\u7684\u5224\u65ad\uff0c\u4ee5\u63d0\u9ad8\u6a21\u578b\u7684\u5224\u65ad\u80fd\u529b\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>\u8fed\u4ee3\u8bad\u7ec3\u8fc7\u7a0b<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li><strong>\u6b65\u9aa41<\/strong>\uff1a\u6f14\u5458\u4e3a\u6bcf\u4e2a\u63d0\u793a\u751f\u6210\u591a\u4e2a\u54cd\u5e94\u53d8\u4f53\u3002<\/li>\n\n\n\n<li><strong>\u6b65\u9aa42<\/strong>\uff1a\u8bc4\u5ba1\u4f7f\u7528\u9884\u5b9a\u4e49\u7684\u6807\u51c6\u8bc4\u4f30\u6bcf\u4e2a\u54cd\u5e94\u5e76\u7ed9\u4e88\u5206\u6570\u3002<\/li>\n\n\n\n<li><strong>\u6b65\u9aa43<\/strong>\uff1a\u5143\u8bc4\u5ba1\u8bc4\u4f30\u8bc4\u5ba1\u7684\u5224\u65ad\uff0c\u521b\u5efa\u5224\u65ad\u7684\u504f\u597d\u5bf9\u7528\u4e8e\u8bad\u7ec3\u3002<\/li>\n\n\n\n<li><strong>\u6b65\u9aa44<\/strong>\uff1a\u5e94\u7528\u504f\u597d\u4f18\u5316\u4ee5\u6539\u8fdb\u6f14\u5458\u548c\u8bc4\u5ba1\u7684\u89d2\u8272\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>\u957f\u5ea6\u63a7\u5236\u673a\u5236<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li>\u65e8\u5728\u51cf\u8f7b\u8bc4\u5ba1\u504f\u5411\u4e8e\u8f83\u957f\u54cd\u5e94\u7684\u503e\u5411\uff0c\u901a\u8fc7\u5c06\u8bc4\u5ba1\u5206\u6570\u4e0e\u957f\u5ea6\u4fe1\u606f\u7ed3\u5408\uff0c\u4ee5\u786e\u4fdd\u5728\u5206\u6570\u63a5\u8fd1\u65f6\u9009\u62e9\u8f83\u77ed\u7684\u9ad8\u8d28\u91cf\u54cd\u5e94\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">\u4e09\u3001\u5b9e\u9a8c<\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u8bbe\u7f6e<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li>\u521d\u59cb\u6a21\u578b\u4e3aLlama-3-8B-Instruct\uff0c\u4e00\u4e2a\u7ecf\u8fc7\u6307\u4ee4\u5fae\u8c03\u7684\u6a21\u578b\u3002<\/li>\n\n\n\n<li>\u9664\u4e86\u521d\u59cb\u6a21\u578b\u5916\uff0c\u4e0d\u4f7f\u7528\u989d\u5916\u7684\u4eba\u7c7b\u76d1\u7763\u8bad\u7ec3\u6570\u636e\u3002<\/li>\n\n\n\n<li>\u6a21\u578b\u7ecf\u5386\u591a\u6b21\u5143\u5956\u52b1\u8bad\u7ec3\uff0c\u91cd\u70b9\u5728\u6f14\u5458\u548c\u8bc4\u5ba1\u89d2\u8272\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>\u8bc4\u4f30\u6307\u6807<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li><strong>\u6307\u4ee4\u9075\u5faa<\/strong>\uff1a\u5728AlpacaEval 2\u3001Arena-Hard\u548cMT-Bench\u7b49\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8fdb\u884c\u8bc4\u4f30\u3002<\/li>\n\n\n\n<li><strong>\u5956\u52b1\u5efa\u6a21<\/strong>\uff1a\u6d4b\u91cf\u6a21\u578b\u751f\u6210\u7684\u5206\u6570\u4e0e\u4eba\u7c7b\u504f\u597d\u7684\u76f8\u5173\u6027\uff0c\u4ee5\u53ca\u4e0e\u5f3a\u5927AI\u8bc4\u5ba1\u7684\u76f8\u5173\u6027\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><strong>\u7ed3\u679c<\/strong>\uff1a\n<ul class=\"wp-block-list\">\n<li>\u5728AlpacaEval 2\u548cArena-Hard\u4e0a\u7684\u80dc\u7387\u663e\u8457\u63d0\u9ad8\u3002<\/li>\n\n\n\n<li>\u589e\u5f3a\u4e86\u4e0e\u4eba\u7c7b\u5224\u65ad\u548c\u5f3a\u5927AI\u8bc4\u5ba1\uff08\u5982GPT-4\uff09\u7684\u76f8\u5173\u6027\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">\u56db\u3001\u4e3b\u8981\u53d1\u73b0<\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u5143\u5956\u52b1<\/strong>\u663e\u8457\u589e\u5f3a\u4e86\u6a21\u578b\u7684\u6f14\u5458\u548c\u8bc4\u5ba1\u6280\u80fd\u3002<\/li>\n\n\n\n<li><strong>\u957f\u5ea6\u63a7\u5236\u673a\u5236<\/strong>\u6709\u6548\u9632\u6b62\u8bad\u7ec3\u8fed\u4ee3\u8fc7\u7a0b\u4e2d\u54cd\u5e94\u957f\u5ea6\u7684\u7206\u70b8\u3002<\/li>\n\n\n\n<li><strong>\u8fed\u4ee3\u8bad\u7ec3<\/strong>\uff1a\u901a\u8fc7\u591a\u6b21\u8fed\u4ee3\u6301\u7eed\u6539\u8fdb\uff0c\u6bcf\u6b21\u8fed\u4ee3\u90fd\u6709\u663e\u8457\u7684\u6027\u80fd\u63d0\u5347\u3002<\/li>\n<\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">\u4e94\u3001\u5c40\u9650\u6027<\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>5\u5206\u8bc4\u5ba1\u7cfb\u7edf<\/strong>\uff1a\u7531\u4e8e\u54cd\u5e94\u4e4b\u95f4\u7684\u8d28\u91cf\u5dee\u5f02\u6700\u5c0f\uff0c\u5f80\u5f80\u5bfc\u81f4\u5e73\u5c40\uff0c\u9700\u8981\u4ed4\u7ec6\u5e73\u5747\u591a\u4e2a\u5224\u65ad\u4ee5\u533a\u5206\u5b83\u4eec\u3002<\/li>\n\n\n\n<li><strong>\u5206\u6570\u9971\u548c<\/strong>\uff1a\u968f\u7740\u8bad\u7ec3\u7684\u8fdb\u884c\uff0c\u54cd\u5e94\u8d8a\u6765\u8d8a\u63a5\u8fd1\u6700\u9ad8\u5206\uff0c\u5bfc\u81f4\u8fdb\u4e00\u6b65\u6539\u8fdb\u96be\u4ee5\u68c0\u6d4b\u3002<\/li>\n\n\n\n<li><strong>\u4f4d\u7f6e\u504f\u5dee<\/strong>\uff1a\u5143\u8bc4\u5ba1\u8bc4\u4f30\u4e2d\u7684\u6301\u7eed\u504f\u5dee\u3002<\/li>\n<\/ol>\n\n\n\n<h4 class=\"wp-block-heading\">\u516d\u3001\u7ed3\u8bba<\/h4>\n\n\n\n<p>\u5143\u5956\u52b1\u673a\u5236\u663e\u8457\u63d0\u9ad8\u4e86LLMs\u7684\u5224\u65ad\u80fd\u529b\uff0c\u589e\u5f3a\u4e86\u5b83\u4eec\u5728\u6ca1\u6709\u989d\u5916\u4eba\u7c7b\u53cd\u9988\u7684\u60c5\u51b5\u4e0b\u9075\u5faa\u6307\u4ee4\u7684\u80fd\u529b\u3002\u8fd9\u79cd\u65b9\u6cd5\u4e3a\u5b9e\u73b0AI\u6a21\u578b\u7684\u8d85\u5bf9\u9f50\uff08super alignment\uff09\u63d0\u4f9b\u4e86\u4e00\u4e2a\u6709\u524d\u9014\u7684\u65b9\u5411\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">\u4e03\u3001\u8fdb\u4e00\u6b65\u7814\u7a76<\/h3>\n\n\n\n<ol class=\"wp-block-list\">\n<li><strong>\u63a2\u7d22\u66f4\u7ec6\u81f4\u7684\u8bc4\u5206\u7cfb\u7edf<\/strong>\uff0c\u4ee5\u66f4\u597d\u5730\u533a\u5206\u54cd\u5e94\u5e76\u907f\u514d\u5206\u6570\u9971\u548c\u3002<\/li>\n\n\n\n<li><strong>\u51cf\u8f7b\u5143\u8bc4\u5ba1\u4e2d\u7684\u4f4d\u7f6e\u504f\u5dee<\/strong>\uff0c\u4ee5\u63d0\u9ad8\u8bad\u7ec3\u6548\u679c\u3002<\/li>\n\n\n\n<li><strong>\u5728\u66f4\u5e7f\u6cdb\u7684\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8bc4\u4f30\u8be5\u65b9\u6cd5<\/strong>\uff0c\u4ee5\u786e\u8ba4\u5176\u901a\u7528\u6027\u548c\u9c81\u68d2\u6027\u3002<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>\u8bba\u6587Meta-Rewarding Language Models: Self-Improving Alignm [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"set","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[8],"tags":[39,77,86],"class_list":["post-4211","post","type-post","status-publish","format-standard","hentry","category-tech","tag-ai","tag-meta","tag-llm"],"rttpg_featured_image_url":null,"rttpg_author":{"display_name":"NullThought","author_link":"https:\/\/nullthought.net\/?author=1"},"rttpg_comment":0,"rttpg_category":"<a href=\"https:\/\/nullthought.net\/?cat=8\" rel=\"category\">Tech<\/a>","rttpg_excerpt":"\u8bba\u6587Meta-Rewarding Language Models: Self-Improving Alignm&hellip;","_links":{"self":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4211","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4211"}],"version-history":[{"count":2,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4211\/revisions"}],"predecessor-version":[{"id":4215,"href":"https:\/\/nullthought.net\/index.php?rest_route=\/wp\/v2\/posts\/4211\/revisions\/4215"}],"wp:attachment":[{"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4211"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4211"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nullthought.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4211"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}