{"id":11176,"date":"2026-08-18T16:50:57","date_gmt":"2026-08-18T08:50:57","guid":{"rendered":"https:\/\/infernews.com\/blog\/agentifying-the-evaluation-of-visual-worlds\/"},"modified":"2026-08-18T17:00:10","modified_gmt":"2026-08-18T09:00:10","slug":"agentifying-the-evaluation-of-visual-worlds","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/agentifying-the-evaluation-of-visual-worlds\/","title":{"rendered":"HarnessEval-W \u8b93\u4e16\u754c\u6a21\u578b\u8a55\u6e2c\u7559\u4e0b\u53ef\u6838\u67e5\u63a8\u7406\u93c8"},"content":{"rendered":"\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/08\/pasted-5da2b18b8496.jpg\" alt=\"HarnessEval logo\"\/><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u7576\u4e16\u754c\u6a21\u578b\u751f\u6210\u4e00\u6bb5\u4e92\u52d5\u5834\u666f\uff0c\u771f\u6b63\u96e3\u5224\u65b7\u7684\u5f80\u5f80\u4e0d\u662f\u756b\u9762\u662f\u5426\u6f02\u4eae\uff0c\u800c\u662f\u7269\u7406\u3001\u56e0\u679c\u548c\u4e16\u754c\u72c0\u614b\u6709\u6c92\u6709\u96a8\u884c\u52d5\u6b63\u78ba\u6f14\u8b8a\u3002HarnessEval-W \u5c6c\u65bc\u958b\u6e90\u7684 agentified benchmark\uff0c\u5be6\u969b\u8655\u7406\u7684\u662f\u5982\u4f55\u628a\u9019\u4e9b\u9700\u8981\u7406\u89e3\u4e0a\u4e0b\u6587\u7684\u8a55\u6e2c\u5de5\u4f5c\uff0c\u8f49\u5316\u6210\u53ef\u6aa2\u67e5\u3001\u53ef\u91cd\u73fe\u7684\u5224\u65b7\u6d41\u7a0b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5b83\u501f\u7528 Large Language Model\uff08LLM\uff09\u751f\u614b\u5e38\u898b\u7684 harness \u601d\u8def\uff0c\u5148\u6309\u6bcf\u500b evaluation case \u7684\u521d\u59cb\u4e16\u754c\u3001action \u548c probe intent \u5206\u89e3\u554f\u984c\uff0c\u518d\u7531\u5177\u5099\u4e0d\u540c\u5de5\u5177\u8207\u4e0a\u4e0b\u6587\u7684 specialized sub-agents \u5206\u5de5\u5224\u65b7\uff0c\u6700\u5f8c\u7531 parent agent \u9a57\u8b49\u8b49\u64da\u4e26\u6574\u5408 verdict\u3002\u6bcf\u6b21\u8a55\u6e2c\u90fd\u6703\u7559\u4e0b evidence tree\uff0c\u8a18\u9304\u6e2c\u8a66\u5167\u5bb9\u3001\u8996\u89ba\u4f9d\u64da\u548c\u5b8c\u6574\u63a8\u7406\u93c8\uff0c\u7814\u7a76\u5718\u968a\u53ef\u4ee5\u8ffd\u67e5\u5206\u6578\u7531\u4f55\u800c\u4f86\uff0c\u800c\u4e0d\u662f\u53ea\u63a5\u53d7\u4e00\u500b\u96e3\u4ee5\u89e3\u91cb\u7684 scalar score\u3002<\/p>\n\n\n<div\n    \tclass=\"align wp-block-vpb-video\"    id='vpbpVideoPlayer-1'\n    data-attributes='{&quot;source&quot;:&quot;https:\\\/\\\/mirros-lab.github.io\\\/HarnessEval-W\\\/assets\\\/gallery-cases\\\/intentional-robot-lab\\\/videos\\\/seedance-2.0-standard-480p-4s-silent.mp4&quot;,&quot;repeat&quot;:true,&quot;autoplay&quot;:true,&quot;muted&quot;:true,&quot;align&quot;:&quot;&quot;,&quot;poster&quot;:&quot;&quot;,&quot;controls&quot;:{&quot;play-large&quot;:true,&quot;restart&quot;:false,&quot;rewind&quot;:true,&quot;play&quot;:true,&quot;fast-forward&quot;:true,&quot;progress&quot;:true,&quot;current-time&quot;:true,&quot;duration&quot;:false,&quot;mute&quot;:true,&quot;volume&quot;:true,&quot;pip&quot;:false,&quot;airplay&quot;:false,&quot;captions&quot;:false,&quot;settings&quot;:true,&quot;download&quot;:false,&quot;fullscreen&quot;:true},&quot;captions&quot;:[],&quot;ratio&quot;:&quot;16:9&quot;,&quot;accentColor&quot;:&quot;&quot;,&quot;lazyLoad&quot;:false,&quot;width&quot;:&quot;100%&quot;,&quot;radius&quot;:&quot;0px&quot;,&quot;resetOnEnd&quot;:false,&quot;autoHideControl&quot;:true,&quot;isSetup&quot;:false}'\n><\/div>\n\n\n<ul class=\"wp-block-list\">\n<li>\u652f\u63f4 330 \u500b evaluation cases\u30015,940 \u6b21 scored rollouts<\/li>\n\n\n\n<li>\u4ee5 11 \u500b specialized evaluation skills \u8986\u84cb\u54c1\u8cea\u3001\u8f49\u63db\u8207\u6301\u7e8c\u6027<\/li>\n\n\n\n<li>\u6db5\u84cb 18 \u500b world models\uff0c\u4e26\u63d0\u4f9b leaderboard \u8207\u53ef\u57f7\u884c\u8a55\u6e2c\u7a0b\u5f0f<\/li>\n\n\n\n<li>\u5305\u542b harnesseval-metrics\uff0c\u4ee5\u53ca physical-plausibility backend harnesseval-pavrm<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">HarnessEval-W \u5728 intentional transition \u4e0a\u8207\u4eba\u985e Bradley\u2013Terry ranking \u7684 Spearman correlation \u70ba 0.93\uff1bphysical transition \u7684 pairwise accuracy \u9054 71.7%\uff0c\u9ad8\u65bc WBench \u7684 31.9%\u3002\u91cd\u8907\u8a55\u6e2c\u7684\u7d50\u679c\u7bc4\u570d\u4ea6\u8f03 WBench \u7a84 4.9 \u500d\uff0c\u4f46\u9019\u4e9b\u6578\u5b57\u4ecd\u53d6\u6c7a\u65bc\u6848\u4f8b\u8a2d\u8a08\u3001\u6280\u80fd\u8def\u7531\u548c\u4ee3\u7406\u5224\u65b7\u54c1\u8cea\uff0c\u4e0d\u80fd\u76f4\u63a5\u8996\u70ba\u6240\u6709\u4e16\u754c\u6a21\u578b\u4efb\u52d9\u7684\u901a\u7528\u6392\u540d\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u9805\u76ee\u63d0\u4f9b\u56fa\u5b9a plans\u3001benchmark \u8cc7\u6e90\u3001metric backends \u548c\u53ef\u5ef6\u4f38\u7684 skill library\uff1b\u5718\u968a\u53ef\u4ee5\u5148\u91cd\u73fe\u65e2\u6709 cases\uff0c\u518d\u63d0\u4ea4\u65b0\u4e16\u754c\u3001\u65b0 action\u3001\u65b0 probe family \u6216\u65b0 skill\u3002\u5b83\u8f03\u9069\u5408\u7814\u7a76\u4e16\u754c\u6a21\u578b\u3001\u5efa\u7acb\u53ef\u5be9\u8a08\u8a55\u6e2c\u6d41\u7a0b\uff0c\u6216\u9700\u8981\u6bd4\u8f03\u7269\u7406\u5408\u7406\u6027\u8207\u72c0\u614b\u6301\u7e8c\u6027\u7684\u5718\u968a\uff1b\u5c0d\u53ea\u60f3\u5feb\u901f\u53d6\u5f97\u55ae\u4e00\u5206\u6578\u7684\u6e2c\u8a66\u6d41\u7a0b\u800c\u8a00\uff0c\u5b8c\u6574\u8b49\u64da\u6a39\u6703\u589e\u52a0\u904b\u7b97\u8207\u6574\u5408\u6210\u672c\uff0c\u4f46\u63db\u4f86\u66f4\u6e05\u695a\u7684\u932f\u8aa4\u5b9a\u4f4d\u80fd\u529b\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/mirros-lab.github.io\/HarnessEval-W\/\" rel=\"noopener noreferrer\"><strong>\u9805\u76ee\u4e3b\u9801<\/strong><\/a> \u00b7 <a href=\"https:\/\/github.com\/MirroS-Lab\/HarnessEval-W\" rel=\"noopener noreferrer\"><strong>GitHub<\/strong><\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u4e16\u754c\u6a21\u578b\u8a55\u6e2c\u4e0d\u518d\u53ea\u7d66\u4e00\u500b\u5206\u6578\uff0cHarnessEval-W\u628a\u6bcf\u6b21\u5224\u65b7\u62c6\u6210\u53ef\u8ffd\u6eaf\u7684\u8b49\u64da\u6a39\uff0c\u8b93\u7269\u7406\u8207\u56e0\u679c\u932f\u8aa4\u66f4\u5bb9\u6613\u6838\u5c0d\u3002<\/p>\n","protected":false},"author":8,"featured_media":11175,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,116,150,186,195,199,201],"tags":[],"class_list":["post-11176","post","type-post","status-publish","format-standard","hentry","category-133","category-agentic","category-150","category-186","category-195","category-dataset-","category-skill-"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11176","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=11176"}],"version-history":[{"count":3,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11176\/revisions"}],"predecessor-version":[{"id":11181,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/11176\/revisions\/11181"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/11175"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=11176"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=11176"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=11176"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}