{"id":29696,"date":"2025-04-01T12:49:58","date_gmt":"2025-04-01T04:49:58","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=29696"},"modified":"2025-04-01T12:49:58","modified_gmt":"2025-04-01T04:49:58","slug":"eq-bench-ruhepingai","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/de\/eq-bench-ruhepingai\/","title":{"rendered":"EQ-Bench \u5982\u4f55\u8bc4\u4f30\u5927\u578b\u8bed\u8a00\u6a21\u578b\u7684\u60c5\u5546\u4e0e\u521b\u9020\u529b"},"content":{"rendered":"<p>\u968f\u7740\u5927\u578b\u8bed\u8a00\u6a21\u578b (LLM) \u80fd\u529b\u7684\u98de\u901f\u53d1\u5c55\uff0c\u4f20\u7edf\u7684\u57fa\u51c6\u6d4b\u8bd5\uff0c\u5982\u00a0<code>MMLU<\/code>\uff0c\u5728\u533a\u5206\u9876\u5c16\u6a21\u578b\u65b9\u9762\u9010\u6e10\u663e\u73b0\u51fa\u5c40\u9650\u6027\u3002\u4ec5\u4ec5\u4f9d\u9760\u77e5\u8bc6\u95ee\u7b54\u6216\u6807\u51c6\u5316\u6d4b\u8bd5\uff0c\u5df2\u96be\u4ee5\u5168\u9762\u8861\u91cf\u6a21\u578b\u5728\u771f\u5b9e\u4e16\u754c\u4ea4\u4e92\u4e2d\u81f3\u5173\u91cd\u8981\u7684\u7ec6\u5fae\u80fd\u529b\uff0c\u4f8b\u5982\u60c5\u5546\u3001\u521b\u9020\u529b\u3001\u5224\u65ad\u529b\u548c\u6c9f\u901a\u6280\u5de7\u3002\u6b63\u662f\u5728\u8fd9\u4e00\u80cc\u666f\u4e0b\uff0c\u4e00\u7cfb\u5217\u540d\u4e3a\u00a0<code>EQ-Bench<\/code>\u00a0\u7684\u65b0\u578b\u8bc4\u4f30\u6846\u67b6\u5e94\u8fd0\u800c\u751f\uff0c\u65e8\u5728\u66f4\u6df1\u5165\u5730\u63a2\u6d4b LLM \u5728\u8fd9\u4e9b\u590d\u6742\u7ef4\u5ea6\u4e0a\u7684\u8868\u73b0\u3002<\/p>\n<p><code>EQ-Bench<\/code>\u00a0\u53ca\u5176\u76f8\u5173\u6d4b\u8bd5\u5957\u4ef6\u63d0\u4f9b\u4e86\u4e00\u4e2a\u591a\u5143\u5316\u7684\u89c6\u89d2\uff0c\u4e0d\u518d\u5c40\u9650\u4e8e\u7b80\u5355\u7684\u5bf9\u9519\u5224\u65ad\uff0c\u800c\u662f\u5173\u6ce8\u6a21\u578b\u5728\u6a21\u62df\u771f\u5b9e\u573a\u666f\u4e2d\u7684\u884c\u4e3a\u548c\u51b3\u7b56\u8d28\u91cf\u3002<\/p>\n<blockquote><p><a href=\"https:\/\/www.kdjingpai.com\/de\/deepseek-chatshena\/\">DeepSeek<\/a> \u5728<a href=\"https:\/\/www.kdjingpai.com\/jiemidamoxingbang\/\">\u77e5\u8bc6\u53ec\u56de\u8bc4\u6d4b\u4e2d\u5e7b\u89c9\u4e1b\u751f<\/a>\uff0c\u4f46\u5199\u4f5c\u521b\u610f\u65b9\u9762\u8003\u8651\u6027\u4ef7\u6bd4\u53ef\u4ee5\u8bf4\u5c60\u699c\uff0c\u8fd9\u5e76\u4e0d\u610f\u5916\uff0c\u548c\u5f88\u591a\u4eba\u5b9e\u9645\u4f7f\u7528\u4f53\u9a8c\u4e5f\u57fa\u672c\u4e00\u81f4\u3002<\/p><\/blockquote>\n<p>&nbsp;<\/p>\n<h3>\u63d0\u4ea4\u6e20\u9053\u4e0e\u8981\u6c42<\/h3>\n<p>\u76ee\u524d\uff0c\u8be5\u9879\u76ee\u4ec5\u63a5\u53d7\u53ef\u901a\u8fc7\u00a0<code>HuggingFace<\/code>\u00a0\u516c\u5f00\u83b7\u53d6\u7684\u5f00\u653e\u6743\u91cd\u6a21\u578b\u7684\u63d0\u6d4b\u3002\u63d0\u4ea4\u8005\u9700\u901a\u8fc7\u7535\u5b50\u90ae\u4ef6\u6216 Twitter \u63d0\u4f9b\u6a21\u578b\u94fe\u63a5\u3001\u6700\u4f73\u63d0\u793a\u683c\u5f0f\u4e0e\u751f\u6210\u914d\u7f6e\uff0c\u4ee5\u53ca\u81ea\u884c\u6d4b\u8bd5\u5f97\u5230\u7684\u00a0<code>EQ-Bench<\/code>\u00a0\u5206\u6570\u3002\u9879\u76ee\u65b9\u4f1a\u8fdb\u884c\u7ed3\u679c\u9a8c\u8bc1\u5e76\u66f4\u65b0\u6392\u884c\u699c\u3002\u7531\u4e8e\u9879\u76ee\u4e3a\u81ea\u7b79\u8d44\u91d1\uff0c\u8ba1\u7b97\u8d44\u6e90\u6709\u9650\uff0c\u63d0\u4ea4\u8005\u9700\u7406\u89e3\u9a8c\u8bc1\u8fc7\u7a0b\u53ef\u80fd\u9700\u8981\u65f6\u95f4\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\ud83d\udc99 EQ-Bench 3: \u51b2\u7a81\u8c03\u89e3\u80fd\u529b\u8bc4\u4f30<\/h3>\n<p><code>EQ-Bench 3<\/code>\u00a0\u662f\u4e00\u4e2a\u4e13\u6ce8\u4e8e\u4e3b\u52a8\u51b2\u7a81\u8c03\u89e3\u6280\u5de7\u7684\u60c5\u5546\u57fa\u51c6\u6d4b\u8bd5\u3002\u5b83\u8bc4\u4f30\u8bed\u8a00\u6a21\u578b\u5728\u5145\u6ee1\u6311\u6218\u548c\u591a\u6837\u5316\u7684\u573a\u666f\u4e2d\u5904\u7406\u590d\u6742\u60c5\u611f\u95ee\u9898\u7684\u80fd\u529b\u3002\u8be5\u6d4b\u8bd5\u4e0d\u518d\u50cf\u521d\u4ee3\u00a0<code>EQ-Bench<\/code>\u00a0\u90a3\u6837\u4ec5\u4ec5\u9884\u6d4b\u60c5\u7eea\u72b6\u6001\uff0c\u800c\u662f\u76f4\u63a5\u8003\u5bdf\u6a21\u578b\u7684\u4e3b\u52a8\u60c5\u5546\u5e94\u7528\u80fd\u529b\u3002<\/p>\n<p><strong>\u6d4b\u8bd5\u65b9\u6cd5:<\/strong><\/p>\n<ul>\n<li><strong>\u7ed3\u6784:<\/strong>\u00a0\u6d4b\u8bd5\u5305\u542b\u591a\u8f6e\u5bf9\u8bdd (\u6700\u591a 21 \u8f6e)\uff0c\u7531\u88ab\u6d4b\u6a21\u578b\u626e\u6f14\u51b2\u7a81\u8c03\u89e3\u5458\uff0c\u4e0e\u626e\u6f14\u5ba2\u6237\u6216\u4e89\u8bae\u65b9\u7684\u201c\u6f14\u5458\u201d\u6a21\u578b (\u76ee\u524d\u4f7f\u7528\u00a0<code>gemini-2.0-flash-001<\/code>) \u8fdb\u884c\u4e92\u52a8\u3002\u6bcf\u4e2a\u573a\u666f\u90fd\u5305\u542b\u8be6\u7ec6\u7684\u89d2\u8272\u8bbe\u5b9a\uff0c\u5177\u6709\u7279\u5b9a\u7684\u60c5\u7eea\u72b6\u6001\u548c\u80cc\u666f\u3002<\/li>\n<li><strong>\u8bc4\u4f30\u6807\u51c6:<\/strong>\u00a0\u8bc4\u5206\u4f9d\u636e\u5305\u62ec\uff1a\n<ul>\n<li>\u57fa\u672c\u60c5\u5546\u6280\u80fd (\u8bc6\u522b\u60c5\u7eea\u3001\u8868\u8fbe\u540c\u7406\u5fc3)<\/li>\n<li>\u6cbb\u7597\u6216\u8c03\u89e3\u76f8\u5173\u7684\u4e13\u4e1a\u6280\u80fd<\/li>\n<li>\u907f\u514d\u4e25\u91cd\u7684\u4e13\u4e1a\u5931\u8bef<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u8fd0\u4f5c\u673a\u5236:<\/strong>\u00a0\u57fa\u51c6\u6d4b\u8bd5\u4f7f\u7528\u4e09\u4e2a\u6a21\u578b\uff1a\u88ab\u6d4b\u6a21\u578b\u3001\u6f14\u5458\u6a21\u578b\u548c\u88c1\u5224\u6a21\u578b (<code>Claude-3.7-Sonnet<\/code>)\u3002\u88c1\u5224\u6a21\u578b\u8d1f\u8d23\u5bf9\u88ab\u6d4b\u6a21\u578b\u7684\u8868\u73b0\u8fdb\u884c\u6253\u5206\u3002<\/li>\n<li><strong>\u8bc4\u5206:<\/strong>\u00a0\u6700\u7ec8\u5f97\u5206\u7ed3\u5408\u4e86\u591a\u4e2a\u6280\u80fd\u9886\u57df\u7684\u8bc4\u5206\u4ee5\u53ca\u8bc6\u522b\u51fa\u7684\u5931\u8bef\u53ca\u5176\u4e25\u91cd\u7a0b\u5ea6\u7684\u8ba1\u6570\u3002\u88c1\u5224\u6a21\u578b\u8fd8\u4f1a\u63d0\u4f9b\u5bf9\u5177\u4f53\u9519\u8bef\u7684\u6279\u5224\u6027\u5206\u6790\uff0c\u5c06\u5176\u8bc4\u4e3a\u8f7b\u5fae\u3001\u4e2d\u5ea6\u6216\u4e25\u91cd\uff0c\u4ece\u800c\u7cbe\u786e\u8bc6\u522b\u6a21\u578b\u5728\u771f\u5b9e\u4e13\u4e1a\u5bf9\u8bdd\u4e2d\u7684\u77ed\u677f\u3002<\/li>\n<\/ul>\n<p><strong>\u88c1\u5224\u6a21\u578b\u7684\u81ea\u6211\u504f\u597d\u95ee\u9898:<\/strong><\/p>\n<p>\u4e00\u4e2a\u5e38\u89c1\u7684\u62c5\u5fe7\u662f LLM \u88c1\u5224\u53ef\u80fd\u5b58\u5728\u504f\u89c1\uff0c\u7279\u522b\u662f\u504f\u7231\u81ea\u8eab\u6a21\u578b\u7684\u8f93\u51fa\u3002\u4e3a\u4e86\u63a2\u7a76\u8fd9\u4e00\u73b0\u8c61\uff0c<code>EQ-Bench 3<\/code>\u00a0\u4f7f\u7528\u00a0<code>Claude-3.7-Sonnet<\/code>\u00a0\u548c\u00a0<code>gpt-4o-2024-11-20<\/code>\u00a0\u4e24\u4e2a\u88c1\u5224\u6a21\u578b\u5bf9\u9876\u7ea7\u6a21\u578b\u8fdb\u884c\u4e86\u57fa\u51c6\u6d4b\u8bd5\u3002<\/p>\n<p><img decoding=\"async\" title=\"EQ-Bench \u5982\u4f55\u8bc4\u4f30\u5927\u578b\u8bed\u8a00\u6a21\u578b\u7684\u60c5\u5546\u4e0e\u521b\u9020\u529b-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/d8e2d30ec868b82.png\" alt=\"EQ-Bench \u5982\u4f55\u8bc4\u4f30\u5927\u578b\u8bed\u8a00\u6a21\u578b\u7684\u60c5\u5546\u4e0e\u521b\u9020\u529b-1\" \/><\/p>\n<p>\u7ed3\u679c\u663e\u793a\uff0c\u81f3\u5c11\u5728\u8fd9\u4e24\u4e2a\u88c1\u5224\u6a21\u578b\u4e4b\u95f4\uff0c\u5e76\u672a\u89c2\u5bdf\u5230\u660e\u663e\u7684\u81ea\u6211\u504f\u597d\uff0c\u5b83\u4eec\u7684\u8bc4\u5206\u51e0\u4e4e\u4e00\u81f4\u3002\u8fd9\u4e3a\u4f7f\u7528 LLM \u4f5c\u4e3a\u88c1\u5224\u63d0\u4f9b\u4e86\u4e00\u5b9a\u7684\u4fe1\u5fc3\uff0c\u4f46\u4ecd\u9700\u6ce8\u610f\u6f5c\u5728\u504f\u89c1\u7684\u53ef\u80fd\u6027\u3002<\/p>\n<p>\u76f8\u5173\u4ee3\u7801\u548c\u5b8c\u6574\u6587\u6863\u9884\u8ba1\u5f88\u5feb\u5c06\u5728\u00a0<code>EQ-Bench<\/code>\u00a0\u4ed3\u5e93\u53d1\u5e03\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\ud83d\udc97 EQ-Bench (Legacy): \u60c5\u7eea\u8bc6\u522b\u57fa\u51c6<\/h3>\n<p>\u521d\u4ee3\u00a0<code>EQ-Bench<\/code>\u00a0\u65e8\u5728\u8bc4\u4f30\u8bed\u8a00\u6a21\u578b\u7684\u60c5\u7eea\u667a\u80fd\uff0c\u8fd9\u90e8\u5206\u80fd\u529b\u5bf9\u7528\u6237\u4f53\u9a8c\u81f3\u5173\u91cd\u8981\uff0c\u5e76\u4e14\u672a\u88ab\u5176\u4ed6\u57fa\u51c6\u660e\u786e\u6d4b\u8bd5\u3002\u5176\u72ec\u7279\u7684\u4f18\u52bf\u5728\u4e8e\uff0c\u901a\u8fc7\u5fae\u8c03\u6765\u201c\u5237\u5206\u201d\u76f8\u5bf9\u56f0\u96be\uff0c\u4f7f\u5f97\u6392\u884c\u699c\u66f4\u80fd\u53cd\u6620\u6a21\u578b\u7684\u771f\u5b9e\u80fd\u529b\u3002<\/p>\n<p>\u8be5\u6d4b\u8bd5\u91c7\u7528\u7279\u5b9a\u683c\u5f0f\uff0c\u8981\u6c42\u6a21\u578b\u9605\u8bfb\u5bf9\u8bdd\u5e76\u8bc4\u4f30\u5176\u4e2d\u4e00\u4e2a\u89d2\u8272\u53ef\u80fd\u60c5\u7eea\u53cd\u5e94\u7684\u5f3a\u5ea6 (\u9488\u5bf9 4 \u79cd\u9884\u8bbe\u60c5\u7eea)\u3002\u8bc4\u4f30\u8fc7\u7a0b\u65e0\u9700\u88c1\u5224\u6a21\u578b\u4ecb\u5165\uff0c\u6210\u672c\u8f83\u4f4e (\u4ec5 171 \u4e2a\u95ee\u9898)\uff0c\u4e14\u7ed3\u679c\u4e0e\u4eba\u7c7b\u504f\u597d (<code>Arena ELO<\/code>) \u53ca\u591a\u9886\u57df\u57fa\u51c6 (<code>MMLU<\/code>) \u8868\u73b0\u51fa\u8f83\u5f3a\u7684\u76f8\u5173\u6027\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\ud83e\uddd9 MAGI-Hard: \u9ad8\u96be\u5ea6\u5224\u522b\u6027\u6d4b\u8bd5\u5b50\u96c6<\/h3>\n<p>\u9274\u4e8e\u73b0\u6709 LLM \u57fa\u51c6\u6d4b\u8bd5\u9762\u4e34\u6027\u80fd\u9971\u548c\u3001\u96be\u4ee5\u533a\u5206\u9876\u5c16\u6a21\u578b\u7684\u95ee\u9898\uff0c<code>MAGI-Hard<\/code>\u00a0\u88ab\u63d0\u51fa\u3002\u5b83\u662f\u4e00\u4e2a\u4ece\u00a0<code>MMLU<\/code>\u00a0\u548c\u00a0<code>AGIEval<\/code>\u00a0\u4e2d\u7cbe\u5fc3\u6311\u9009\u51fa\u7684\u5b50\u96c6\uff0c\u65e8\u5728\u589e\u5f3a\u5bf9\u9ad8\u80fd\u529b\u6a21\u578b\u7684\u533a\u5206\u5ea6\u3002\u5f00\u53d1\u8005\u53ef\u4ee5\u5229\u7528\u63d0\u4f9b\u7684\u00a0<code>EleutherAI lm-evaluation-harness<\/code>\u00a0\u5206\u652f\u6765\u8fd0\u884c\u00a0<code>MAGI<\/code>\u00a0\u6d4b\u8bd5\u96c6\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\ud83c\udfa8 Creative Writing v3: \u5347\u7ea7\u7248\u521b\u610f\u5199\u4f5c\u8bc4\u4f30<\/h3>\n<p>\u521b\u610f\u5199\u4f5c\u57fa\u51c6\u6d4b\u8bd5\u8fce\u6765\u4e86 v3 \u7248\u672c\uff0c\u4e3b\u8981\u66f4\u65b0\u5305\u62ec\uff1a<\/p>\n<ul>\n<li><strong>\u65b0\u63d0\u793a\u8bcd:<\/strong>\u00a0\u65e8\u5728\u66f4\u597d\u5730\u8fdb\u884c\u533a\u5206\u5ea6\u548c\u201c\u6c1b\u56f4\u611f\u201d\u68c0\u67e5\u3002<\/li>\n<li><strong>\u6df7\u5408\u8bc4\u5206\u7cfb\u7edf:<\/strong>\u00a0\u7ed3\u5408\u4e86\u8bc4\u5206\u7ec6\u5219 (Rubric) \u548c\u00a0<code>Elo<\/code>\u00a0\u8bc4\u5206\u7cfb\u7edf\uff0c\u663e\u8457\u63d0\u5347\u4e86\u9876\u5c16\u6a21\u578b\u7684\u533a\u5206\u5ea6\u548c\u6d4b\u8bd5\u7684\u52a8\u6001\u8303\u56f4\u3002<\/li>\n<li><strong>\u6539\u8fdb\u7684\u6837\u672c\u8f93\u51fa\u9875\u9762\u3002<\/strong><\/li>\n<\/ul>\n<p><strong>\u57fa\u51c6\u6d4b\u8bd5\u6d41\u7a0b:<\/strong><\/p>\n<ol>\n<li>\u4f7f\u7528 32 \u4e2a\u5199\u4f5c\u63d0\u793a\u8fd0\u884c\u6a21\u578b 3 \u6b21\u8fed\u4ee3 (\u5171 96 \u9879)\uff0c\u6e29\u5ea6\u8bbe\u7f6e\u4e3a 0.7\uff0cmin_p \u4e3a 0.1\u3002<\/li>\n<li>\u4f7f\u7528\u00a0<code><a href=\"https:\/\/www.kdjingpai.com\/de\/claudeanquanfubai\/\">Claude<\/a> 3.7 Sonnet<\/code>\u00a0\u4f5c\u4e3a\u88c1\u5224\uff0c\u6839\u636e\u5168\u9762\u7684\u8bc4\u5206\u7ec6\u5219\u5bf9\u8f93\u51fa\u8fdb\u884c\u8bc4\u5206\u3002<\/li>\n<li>\u57fa\u4e8e\u7ec6\u5219\u5f97\u5206\u63a8\u65ad\u6a21\u578b\u7684\u521d\u59cb\u00a0<code>Elo<\/code>\u00a0\u8bc4\u5206\u3002<\/li>\n<li>\u4e0e\u6392\u884c\u699c\u4e0a\u76f8\u90bb\u7684\u6a21\u578b\u8fdb\u884c\u6210\u5bf9\u6bd4\u8f83 (\u7a00\u758f\u62bd\u6837)\uff0c\u6839\u636e\u591a\u4e2a\u6807\u51c6\u8bc4\u5206\uff0c\u80dc\u8005\u83b7\u5f97\u6700\u591a 5 \u4e2a &#8220;+&#8221;\u3002<\/li>\n<li>\u4f7f\u7528\u00a0<code>Glicko<\/code>\u00a0\u8bc4\u5206\u7cfb\u7edf (\u4fee\u6539\u540e\u8003\u8651 &#8220;+&#8221; \u6570\u91cf\u7684\u80dc\u8d1f\u5e45\u5ea6) \u8ba1\u7b97\u00a0<code>Elo<\/code>\u00a0\u5206\u6570\uff0c\u5faa\u73af\u76f4\u81f3\u6392\u540d\u7a33\u5b9a\u3002<\/li>\n<li>\u4e0e\u6700\u7ec8\u786e\u5b9a\u7684\u76f8\u90bb\u6a21\u578b\u8fdb\u884c\u5168\u9762\u6bd4\u8f83\uff0c\u8ba1\u7b97\u6700\u7ec8\u7684\u6392\u884c\u699c\u00a0<code>Elo<\/code>\u3002<\/li>\n<\/ol>\n<p><strong>\u8bc4\u5206\u7ec6\u5219\u5f97\u5206 vs Elo \u5f97\u5206:<\/strong><\/p>\n<p>\u4e24\u79cd\u8bc4\u5206\u65b9\u5f0f\u5e76\u5b58\u3002\u7ec6\u5219\u8bc4\u5206\u662f\u88c1\u5224\u72ec\u7acb\u8bc4\u4f30\u5355\u4e2a\u6a21\u578b\u8f93\u51fa\u7684\u7ed3\u679c\uff1b<code>Elo<\/code>\u00a0\u8bc4\u5206\u5219\u662f\u901a\u8fc7\u6a21\u578b\u95f4\u7684\u76f4\u63a5\u4e24\u4e24\u5bf9\u6297\u4ea7\u751f\u7684\u76f8\u5bf9\u6392\u540d\u3002\u4e24\u8005\u7684\u5dee\u5f02\u6e90\u4e8e\u8bc4\u4f30\u65b9\u5f0f\uff1a\u6210\u5bf9\u6bd4\u8f83\u4f7f\u88c1\u5224\u66f4\u5bb9\u6613\u53d1\u73b0\u7ec6\u5fae\u5dee\u522b\uff0c\u4f46\u4e5f\u53ef\u80fd\u5f15\u5165\u4e0d\u540c\u504f\u89c1\u3002\u8bc4\u5206\u6807\u51c6\u5728\u4e24\u79cd\u6a21\u5f0f\u4e0b\u4e5f\u6709\u6240\u4e0d\u540c\u3002<\/p>\n<p><strong>\u5206\u6570\u6807\u51c6\u5316:<\/strong><\/p>\n<p>\u4e3a counteract\u00a0<code>Elo<\/code>\u00a0\u7cfb\u7edf\u4e2d\u65b0\u6a21\u578b\u52a0\u5165\u5bfc\u81f4\u6574\u4f53\u5206\u6570\u6f02\u79fb\u7684\u95ee\u9898\uff0c\u6392\u884c\u699c\u5c06\u00a0<code><a href=\"https:\/\/www.kdjingpai.com\/de\/deepseek-r1nenglixiang\/\">DeepSeek-R1<\/a><\/code>\u00a0\u7684\u5206\u6570\u951a\u5b9a\u4e3a 1500\uff0c<code>ministral-3b<\/code>\u00a0\u951a\u5b9a\u4e3a 200\u3002<\/p>\n<p><strong>\u57fa\u51c6\u6d4b\u8bd5\u54f2\u5b66:<\/strong><\/p>\n<p>\u53ef\u9760\u5730\u3001\u7b26\u5408\u4eba\u7c7b\u504f\u597d\u5730\u8bc4\u5224\u521b\u610f\u5199\u4f5c\u975e\u5e38\u56f0\u96be\u3002v3 \u7248\u672c\u901a\u8fc7\u5f15\u5165\u6210\u5bf9\u6bd4\u8f83\u548c\u00a0<code>Elo<\/code>\u00a0\u6392\u540d\u7cfb\u7edf\u6765\u63d0\u5347\u533a\u5206\u5ea6\u3002\u63d0\u793a\u8bcd\u7ecf\u8fc7\u7cbe\u5fc3\u7b5b\u9009\uff0c\u65e8\u5728\u6311\u6218\u6a21\u578b\u5f31\u70b9\uff0c\u4ece\u800c\u4e3a\u88c1\u5224\u521b\u9020\u66f4\u9661\u5ced\u7684\u8bc4\u4f30\u68af\u5ea6\u3002\u6d4b\u8bd5\u5185\u5bb9\u5305\u542b\u5e7d\u9ed8\u3001\u6d6a\u6f2b\u3001\u7a7a\u95f4\u610f\u8bc6\u3001\u975e\u4f20\u7edf\u7b2c\u4e00\u4eba\u79f0\u89c6\u89d2\u7b49 LLM \u901a\u5e38\u96be\u4ee5\u8fbe\u5230\u4eba\u7c7b\u6c34\u5e73\u7684\u65b9\u9762\u3002<\/p>\n<p><strong>\u6210\u672c\u4e0e\u504f\u89c1\u7f13\u89e3:<\/strong><\/p>\n<p>\u5c3d\u7ba1\u91c7\u7528\u4e86\u6df7\u5408\u8bc4\u5206\u7cfb\u7edf\uff0c\u4f46\u8fd0\u884c\u4e00\u6b21\u8bc4\u4f30\u4ecd\u9700\u7ea6 10 \u7f8e\u5143\u7684 API \u8d39\u7528\u3002\u6210\u5bf9\u6bd4\u8f83\u5f15\u5165\u4e86\u65b0\u7684\u504f\u89c1\u6311\u6218\uff0c\u9879\u76ee\u65b9\u5c1d\u8bd5\u7f13\u89e3\u4e86\uff1a<\/p>\n<ul>\n<li><strong>\u957f\u5ea6\u504f\u89c1:<\/strong>\u00a0\u901a\u8fc7\u5c06\u8f93\u51fa\u622a\u65ad\u4e3a 4000 \u5b57\u7b26\u6765\u63a7\u5236\u3002<\/li>\n<li><strong>\u4f4d\u7f6e\u504f\u89c1:<\/strong>\u00a0\u901a\u8fc7\u53cc\u5411\u8bc4\u4f30\u5e76\u53d6\u5e73\u5747\u503c\u6765\u7f13\u89e3\u3002<\/li>\n<li><strong>\u590d\u6742\u5197\u8bcd\u504f\u89c1:<\/strong>\u00a0\u8bc4\u5206\u6807\u51c6\u5305\u542b\u5bf9\u8fc7\u5ea6\u70ab\u6280\u8bcd\u6c47\u7684\u60e9\u7f5a\u3002<\/li>\n<li><strong>\u8bd7\u610f\u6666\u6da9\u504f\u89c1:<\/strong>\u00a0\u8bc4\u5206\u6807\u51c6\u8bd5\u56fe\u60e9\u7f5a\u8fc7\u5ea6\u8bd7\u610f\u5316\u548c\u4e0d\u8fde\u8d2f\u7684\u6563\u6587\u3002<\/li>\n<\/ul>\n<p><strong>\u672a\u63a7\u5236\u7684\u504f\u89c1:<\/strong><\/p>\n<ul>\n<li><strong>\u81ea\u6211\u504f\u89c1:<\/strong>\u00a0\u672a\u63a7\u5236\u88c1\u5224\u53ef\u80fd\u504f\u7231\u81ea\u8eab\u8f93\u51fa\u3002<\/li>\n<li><strong>\u79ef\u6781\u6027\u504f\u89c1:<\/strong>\u00a0\u65b9\u5411\u548c\u5f71\u54cd\u5c1a\u4e0d\u660e\u786e\u3002<\/li>\n<li><strong>\u4f4e\u4fd7\u5185\u5bb9\u504f\u89c1 (Smut Bias):<\/strong>\u00a0\u88c1\u5224\u503e\u5411\u4e8e\u4e25\u5389\u60e9\u7f5a\u504f\u5411\u60c5\u8272\u7684\u5185\u5bb9\u3002<\/li>\n<li><strong>\u98ce\u683c\u4e0e\u5185\u5bb9\u504f\u89c1:<\/strong>\u00a0\u88c1\u5224\u7684\u504f\u597d\u53ef\u80fd\u4e0e\u7528\u6237\u6216\u666e\u904d\u4eba\u7c7b\u504f\u597d\u4e0d\u540c\u3002<\/li>\n<li><strong>\u5957\u8bdd\u504f\u89c1 (Slop Bias):<\/strong>\u00a0\u88c1\u5224\u53ef\u80fd\u504f\u7231 LLM \u5e38\u7528\u7684\u67d0\u4e9b\u5957\u8bdd\u6216\u98ce\u683c\u3002<\/li>\n<\/ul>\n<p><strong>\u6210\u5bf9\u8bc4\u5224\u63d0\u793a\u6838\u5fc3\u6807\u51c6:<\/strong>\u00a0\u89d2\u8272\u771f\u5b9e\u6027\u3001\u8da3\u5473\u4e0e\u539f\u521b\u6027\u3001\u5199\u4f5c\u8d28\u91cf\u3001\u8fde\u8d2f\u6027\u3001\u6307\u4ee4\u9075\u5faa\u5ea6\u3001\u4e16\u754c\u4e0e\u6c1b\u56f4\u6784\u5efa\u3001\u907f\u514d\u9648\u8bcd\u6ee5\u8c03\u3001\u907f\u514d\u534e\u4e3d\u5197\u8bcd\u3001\u907f\u514d\u8fc7\u5ea6\u9690\u55bb\u3002<\/p>\n<p><strong>\u5c40\u9650\u6027:<\/strong><\/p>\n<p>\u8bc4\u4f30\u7ed3\u679c\u4ec5\u4e3a\u5199\u4f5c\u80fd\u529b\u7684\u7c97\u7565\u53c2\u8003\u3002\u521b\u610f\u5199\u4f5c\u4e3b\u89c2\u6027\u5f3a\uff0c\u5efa\u8bae\u7ed3\u5408\u9605\u8bfb\u6837\u672c\u8f93\u51fa\u6765\u81ea\u884c\u5224\u65ad\u3002\u8be5\u57fa\u51c6\u5e76\u975e\u89d2\u8272\u626e\u6f14 (RP) \u8bc4\u4f30\uff0c\u4e0d\u6d4b\u8bd5\u591a\u8f6e\u5bf9\u8bdd\uff0c\u4e14\u4ec5\u9650\u82f1\u8bed\u5199\u4f5c\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\ud83c\udfa8 Creative Writing (Legacy v2): \u65e7\u7248\u521b\u610f\u5199\u4f5c\u8bc4\u4f30<\/h3>\n<p>\u65e7\u7248 (v2) \u4f7f\u7528\u00a0<code>Claude 3.5 Sonnet<\/code>\u00a0\u4f5c\u4e3a\u88c1\u5224\uff0c\u8bc4\u4f30\u6a21\u578b\u5728\u4e00\u7cfb\u5217\u63d0\u793a\u4e0b\u7684\u5199\u4f5c\u80fd\u529b\u3002\u5b83\u5f15\u5165\u4e86 &#8220;Vocab Complexity&#8221; (\u8bcd\u6c47\u590d\u6742\u5ea6) \u548c &#8220;GPT-Slop&#8221; (GPT \u5957\u8bdd) \u6307\u6807\uff0c\u5e76\u5141\u8bb8\u7528\u6237\u901a\u8fc7\u6ed1\u5757\u8c03\u6574\u5bf9\u8fd9\u4e24\u9879\u7684\u60e9\u7f5a\u6743\u91cd\u3002\u8be5\u7248\u672c\u91c7\u7528\u8bc4\u5206\u7ec6\u5219\u548c\u53c2\u8003\u8f93\u51fa\u6765\u8bc4\u4f30 24 \u4e2a\u63d0\u793a\uff0c\u8fd0\u884c 10 \u6b21\u8fed\u4ee3\u4ee5\u63d0\u9ad8\u7a33\u5b9a\u6027\u3002\u540c\u6837\uff0c\u5b83\u4e5f\u5c1d\u8bd5\u901a\u8fc7\u8bc4\u5206\u6807\u51c6\u548c\u957f\u5ea6\u63a7\u5236\u6ed1\u5757\u6765\u7f13\u89e3\u957f\u5ea6\u504f\u89c1\uff0c\u4f46\u627f\u8ba4\u504f\u89c1\u53ef\u80fd\u4f9d\u7136\u5b58\u5728\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\u2696\ufe0f Judgemark V2: \u88c1\u5224\u6a21\u578b\u80fd\u529b\u8bc4\u4f30<\/h3>\n<p><code>Judgemark V2<\/code>\u00a0\u8bc4\u4f30\u7684\u662f\u8bed\u8a00\u6a21\u578b\u4f5c\u4e3a\u201c\u88c1\u5224\u201d\u672c\u8eab\u7684\u80fd\u529b\uff0c\u5373\u5b83\u4eec\u6839\u636e\u8be6\u7ec6\u7684\u7ec6\u5219\u5bf9\u521b\u610f\u5199\u4f5c\u8fdb\u884c\u00a0<em>\u6570\u503c\u8bc4\u5206<\/em>\u00a0\u7684\u80fd\u529b\u3002\u8fd9\u6bd4\u7b80\u5355\u7684\u6210\u5bf9\u504f\u597d\u6d4b\u8bd5\u66f4\u590d\u6742\uff0c\u8981\u6c42\u88c1\u5224\u6a21\u578b\u7406\u89e3\u590d\u6742\u6307\u4ee4\uff0c\u5206\u6790\u6587\u672c\uff0c\u5e76\u4e3a\u591a\u8fbe 36 \u4e2a\u6587\u5b66\u8d28\u91cf\u7ef4\u5ea6\u6253\u5206\u3002<\/p>\n<p><strong>V2 \u7248\u672c\u4e3b\u8981\u6539\u8fdb:<\/strong><\/p>\n<ul>\n<li>\u6837\u672c\u91cf\u589e\u52a0 6 \u500d\uff0c\u51cf\u5c11\u8fd0\u884c\u65b9\u5dee\u3002<\/li>\n<li>\u7cbe\u70bc\u7684\u8bc4\u5206\u6307\u6807\uff1a\u8861\u91cf\u533a\u5206\u5ea6 (\u533a\u5206\u5f3a\u5f31\u6587\u672c)\u3001\u7a33\u5b9a\u6027 (\u8de8\u8fd0\u884c\u6392\u540d\u4e00\u81f4\u6027) \u53ca\u4e0e\u4eba\u7c7b\u504f\u597d\u7684\u76f8\u5173\u6027\u3002<\/li>\n<li>\u63d0\u4f9b\u539f\u59cb\u5206\u6570\u548c\u6821\u51c6\u5206\u6570\uff1a\u540e\u8005\u901a\u8fc7\u6807\u51c6\u5316\u5206\u5e03\u4f7f\u4e0d\u540c\u88c1\u5224\u66f4\u5177\u53ef\u6bd4\u6027\u3002<\/li>\n<li>\u5f15\u5165\u6270\u52a8\u7a33\u5b9a\u6027\u6d4b\u8bd5 (temp=0.5, top_k=3) \u6765\u8bc4\u4f30\u7ed3\u679c\u7684\u7a33\u5065\u6027\u3002<\/li>\n<li>\u7b80\u5316\u7684\u72ec\u7acb\u4ee3\u7801\u5e93\u3002<\/li>\n<\/ul>\n<p><strong>\u53ef\u91cd\u590d\u6027\u6d4b\u8bd5:<\/strong>\u00a0\u5bf9\u00a0<code>Llama-3.1-70B-instruct<\/code>\u00a0\u8fdb\u884c 20 \u6b21\u6d4b\u8bd5\uff0c\u663e\u793a\u6700\u7ec8\u5f97\u5206\u6807\u51c6\u5dee\u7ea6\u4e3a 1.0\uff0c\u8868\u660e\u7ed3\u679c\u5177\u6709\u4e00\u5b9a\u7684\u7a33\u5b9a\u6027\u3002<\/p>\n<p><strong>\u8bc4\u5224\u4efb\u52a1:<\/strong>\u00a0\u88c1\u5224\u6a21\u578b\u9700\u9605\u8bfb\u7531 17 \u4e2a\u4e0d\u540c\u6c34\u5e73\u201c\u5199\u624b\u6a21\u578b\u201d\u751f\u6210\u7684\u77ed\u7bc7\u521b\u610f\u4f5c\u54c1\uff0c\u5e76\u6839\u636e\u5305\u542b\u6b63\u53cd\u4e24\u65b9\u9762\u6807\u51c6\u7684\u957f\u63d0\u793a (\u5982\u201c\u7ec6\u81f4\u7684\u89d2\u8272: 0-10\u201d\uff0c\u201c\u77eb\u63c9\u9020\u4f5c: 0-10\u201d\uff0c\u540e\u8005\u5206\u6570\u8d8a\u4f4e\u8d8a\u597d) \u8f93\u51fa\u6570\u503c\u8bc4\u5206\u3002\u6700\u7ec8\u00a0<code>Judgemark<\/code>\u00a0\u5206\u6570\u662f\u57fa\u4e8e\u533a\u5206\u5ea6\u3001\u7a33\u5b9a\u6027\u3001\u4e0e\u4eba\u7c7b\u504f\u597d\u76f8\u5173\u6027\u7b49\u591a\u4e2a\u8ba1\u7b97\u6307\u6807\u7684\u52a0\u6743\u603b\u548c\u3002<\/p>\n<p>\u8fd9\u662f\u4e00\u4e2a\u5bf9 LLM \u6781\u5177\u6311\u6218\u6027\u7684\u4efb\u52a1\uff0c\u56e0\u4e3a\u5b83\u9700\u8981\u7ec6\u81f4\u7684\u6587\u5b66\u8bc4\u8bba\u80fd\u529b\u548c\u9075\u5faa\u591a\u7ef4\u6570\u503c\u8bc4\u5206\u6307\u4ee4\u7684\u80fd\u529b\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\ud83c\udfa4 BuzzBench: \u5e7d\u9ed8\u5206\u6790\u57fa\u51c6<\/h3>\n<p><code>BuzzBench<\/code>\u00a0\u901a\u8fc7\u5206\u6790\u82f1\u56fd\u97f3\u4e50\u95ee\u7b54\u8282\u76ee\u00a0<code>Never Mind The Buzzcocks<\/code>\u00a0\u4e2d\u7684\u5609\u5bbe\u4ecb\u7ecd\u7b11\u8bdd\uff0c\u6765\u8bc4\u4f30 LLM \u7684\u5e7d\u9ed8\u7406\u89e3\u80fd\u529b\u3002\u4efb\u52a1\u4e0d\u4ec5\u8981\u6c42\u89e3\u91ca\u7b11\u8bdd\u7684\u673a\u5236\uff0c\u8fd8\u8981\u9884\u6d4b\u7b11\u8bdd\u5bf9\u89c2\u4f17\u548c\u559c\u5267\u4f5c\u5bb6\u7684\u201c\u597d\u7b11\u7a0b\u5ea6\u201d\u3002<\/p>\n<p><strong>\u6311\u6218\u6027\u8bbe\u8ba1:<\/strong><\/p>\n<ul>\n<li>\u9009\u7528\u8282\u76ee\u7b11\u8bdd\uff0c\u56e0\u5176\u98ce\u683c\u591a\u6837\uff08\u5fae\u5999\u3001\u5192\u9669\u3001\u7c97\u7cd9\u3001\u6666\u6da9\u3001\u660e\u663e\u3001\u5de7\u5999\uff09\uff0c\u589e\u52a0\u4e86 LLM \u7684\u8fa8\u522b\u96be\u5ea6\u3002<\/li>\n<li>\u8981\u6c42\u9884\u6d4b\u201c\u597d\u7b11\u7a0b\u5ea6\u201d\uff0c\u6d89\u53ca\u5bf9\u4eba\u7c7b\u53cd\u5e94\u7684\u5efa\u6a21\u3002<\/li>\n<li>\u4f7f\u7528\u4eba\u7c7b\u64b0\u5199\u7684\u201c\u9ec4\u91d1\u7b54\u6848\u201d\u4f5c\u4e3a\u88c1\u5224\u8bc4\u5206\u7684\u57fa\u51c6\u3002<\/li>\n<\/ul>\n<p>\u8be5\u57fa\u51c6\u65e8\u5728\u6d4b\u8bd5\u7406\u8bba\u5fc3\u667a\u7406\u89e3\u548c\u5bf9\u7b11\u8bdd\u8fd0\u4f5c\u673a\u5236\u7684\u590d\u6742\u8ba4\u77e5\u3002\u88c1\u5224\u6a21\u578b\u9009\u7528\u00a0<code>Claude 3.5 Sonnet<\/code>\uff0c\u56e0\u4e3a\u5b83\u5728\u00a0<code>Judgemark<\/code>\u00a0\u4e0a\u5f97\u5206\u9ad8\uff0c\u4e14\u4f3c\u4e4e\u8f83\u5c11\u504f\u7231\u5197\u957f\u3001\u8fc7\u5ea6\u5206\u6790\u7684\u56de\u7b54\u3002\u540c\u6837\u9700\u8981\u6ce8\u610f\u88c1\u5224\u7684\u6f5c\u5728\u81ea\u6211\u504f\u597d\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>\ud83c\udf0d DiploBench: \u6218\u7565\u8c08\u5224\u8bc4\u4f30\u6846\u67b6 (\u5b9e\u9a8c\u6027)<\/h3>\n<p><code>DiploBench<\/code>\u00a0\u662f\u4e00\u4e2a\u5229\u7528\u7b56\u7565\u68cb\u76d8\u6e38\u620f\u00a0<code>Diplomacy<\/code>\u00a0\u8bc4\u4f30 LLM \u6218\u7565\u8c08\u5224\u80fd\u529b\u7684\u5b9e\u9a8c\u6027\u6846\u67b6\u3002\u88ab\u6d4b\u6a21\u578b\u626e\u6f14\u00a0<code>Austria-Hungary<\/code>\u00a0(\u5965\u5308\u5e1d\u56fd)\uff0c\u4e00\u4e2a\u9700\u8981\u9ad8\u8d85\u8c08\u5224\u6280\u5de7\u548c\u6218\u7565\u89c4\u5212\u7684\u6311\u6218\u6027\u89d2\u8272\u3002\u6a21\u578b\u9700\u4e0e\u5176\u4ed6 AI \u73a9\u5bb6\u6c9f\u901a\u3001\u7ed3\u76df\u3001\u8bc6\u522b\u6b3a\u9a97\u5e76\u505a\u51fa\u6218\u672f\u51b3\u7b56\u3002<\/p>\n<p><strong>\u4e3b\u8981\u7279\u70b9:<\/strong><\/p>\n<ul>\n<li><strong>\u5b8c\u5168\u6c9f\u901a\u6a21\u5f0f (Full-press):<\/strong>\u00a0\u884c\u52a8\u524d\u8fdb\u884c\u591a\u8f6e\u8c08\u5224\u3002<\/li>\n<li><strong>\u591a\u667a\u80fd\u4f53\u73af\u5883:<\/strong>\u00a0\u6bcf\u4e2a\u56fd\u5bb6\u7531\u72ec\u7acb\u7684 LLM \u63a7\u5236\u3002<\/li>\n<li><strong>\u73b0\u5b9e\u5916\u4ea4\u6a21\u62df:<\/strong>\u00a0\u6d4b\u8bd5\u7ed3\u76df\u3001\u8c08\u5224\u3001\u6b3a\u9a97\u8bc6\u522b\u80fd\u529b\u3002<\/li>\n<li><strong>\u6311\u6218\u6027\u5f00\u5c40:<\/strong>\u00a0<code>Austria-Hungary<\/code>\u00a0\u7684\u4e2d\u5fc3\u4f4d\u7f6e\u4f7f\u5176\u8106\u5f31\u4f46\u5177\u6218\u7565\u610f\u4e49\u3002<\/li>\n<\/ul>\n<p>\u6e38\u620f\u6700\u591a\u8fd0\u884c 50 \u56de\u5408\uff0c\u6bcf\u56de\u5408\u884c\u52a8\u524d\u6709 4 \u8f6e\u8c08\u5224\u3002\u8be5\u6846\u67b6\u72ec\u7279\u5730\u6d4b\u8bd5\u4e86 LLM \u5728\u957f\u671f\u6218\u7565\u89c4\u5212\u3001\u591a\u667a\u80fd\u4f53\u8c08\u5224\u3001\u7406\u8bba\u5fc3\u667a\u548c\u6b3a\u9a97\u68c0\u6d4b\u65b9\u9762\u7684\u80fd\u529b\u3002<\/p>\n<p><em>\u6ce8\u610f\uff1a\u7531\u4e8e\u6e38\u620f\u8fd0\u884c\u7ed3\u679c\u65b9\u5dee\u8f83\u5927\uff0c<code>DiploBench<\/code>\u00a0\u76ee\u524d\u4ecd\u662f\u5b9e\u9a8c\u6027\u6846\u67b6\uff0c\u7ed3\u679c\u89e3\u8bfb\u9700\u8c28\u614e\u3002<\/em><\/p>\n<p><strong>\u603b\u7ed3\u601d\u8003<\/strong><\/p>\n<p><code>EQ-Bench<\/code>\u00a0\u7cfb\u5217\u63d0\u4f9b\u4e86\u4e00\u5957\u6709\u4ef7\u503c\u7684\u5de5\u5177\uff0c\u63a8\u52a8 LLM \u8bc4\u4f30\u8d85\u8d8a\u4f20\u7edf\u6307\u6807\uff0c\u8f6c\u5411\u66f4\u7b26\u5408\u4eba\u7c7b\u4ea4\u4e92\u9700\u6c42\u7684\u590d\u6742\u80fd\u529b\u7ef4\u5ea6\u3002\u867d\u7136\u8bf8\u5982\u88c1\u5224\u504f\u89c1\u3001\u8bc4\u4f30\u4e3b\u89c2\u6027\u7b49\u6311\u6218\u4f9d\u7136\u5b58\u5728\uff0c\u4f46\u8fd9\u4e9b\u57fa\u51c6\u7684\u8bbe\u8ba1\u548c\u8fed\u4ee3\uff08\u5982 v2\u3001v3 \u7684\u6539\u8fdb\uff0c\u5bf9\u504f\u89c1\u7684\u660e\u786e\u8ba8\u8bba\u548c\u7f13\u89e3\u5c1d\u8bd5\uff09\u4ee3\u8868\u4e86 LLM \u8bc4\u4f30\u9886\u57df\u5411\u66f4\u6df1\u5c42\u6b21\u3001\u66f4\u7ec6\u81f4\u5316\u65b9\u5411\u53d1\u5c55\u7684\u91cd\u8981\u4e00\u6b65\u3002\u5bf9\u4e8e\u5f00\u53d1\u8005\u548c\u7814\u7a76\u8005\u800c\u8a00\uff0c\u8fd9\u4e9b\u5de5\u5177\u4e0d\u4ec5\u63d0\u4f9b\u4e86\u8861\u91cf\u6a21\u578b\u8fdb\u6b65\u7684\u6807\u5c3a\uff0c\u4e5f\u6307\u660e\u4e86\u672a\u6765\u63d0\u5347\u6a21\u578b\u80fd\u529b\u7684\u5173\u952e\u65b9\u5411\uff1a\u60c5\u5546\u3001\u521b\u9020\u529b\u3001\u5224\u65ad\u529b\u4ee5\u53ca\u5728\u590d\u6742\u793e\u4ea4\u4e92\u52a8\u4e2d\u7684\u6709\u6548\u6c9f\u901a\u3002<\/p>\n<p>&nbsp;<\/p>\n<h3>EQ-Bench 3 \u6392\u884c<\/h3>\n<p><img decoding=\"async\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/04\/3010cc7941486ad.png\" alt=\"\" \/><\/p>\n<p>https:\/\/eqbench.com\/<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u968f\u7740\u5927\u578b\u8bed\u8a00\u6a21\u578b (LLM) \u80fd\u529b\u7684\u98de\u901f\u53d1\u5c55\uff0c\u4f20\u7edf\u7684\u57fa\u51c6\u6d4b\u8bd5\uff0c\u5982\u00a0MMLU\uff0c\u5728\u533a\u5206\u9876\u5c16\u6a21\u578b\u65b9\u9762\u9010\u6e10\u663e\u73b0\u51fa\u5c40\u9650\u6027\u3002\u4ec5\u4ec5\u4f9d\u9760\u77e5\u8bc6\u95ee\u7b54\u6216\u6807\u51c6\u5316\u6d4b\u8bd5\uff0c\u5df2\u96be\u4ee5\u5168\u9762\u8861\u91cf\u6a21\u578b\u5728\u771f\u5b9e\u4e16\u754c\u4ea4\u4e92\u4e2d\u81f3\u5173\u91cd\u8981\u7684\u7ec6\u5fae\u80fd\u529b\uff0c\u4f8b\u5982\u60c5\u5546\u3001\u521b\u9020\u529b\u3001\u5224\u65ad\u529b\u548c\u6c9f\u901a\u6280\u5de7\u3002\u6b63\u662f\u5728\u8fd9&#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[34],"tags":[],"class_list":["post-29696","post","type-post","status-publish","format-standard","hentry","category-knowledge"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/29696","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/comments?post=29696"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/posts\/29696\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/media?parent=29696"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/categories?post=29696"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/de\/wp-json\/wp\/v2\/tags?post=29696"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}