{"id":36856,"date":"2025-01-19T23:00:34","date_gmt":"2025-01-19T15:00:34","guid":{"rendered":"https:\/\/17aitech.com\/?p=36856"},"modified":"2025-01-19T23:00:34","modified_gmt":"2025-01-19T15:00:34","slug":"scaling-law-%e6%92%9e%e5%a2%99%ef%bc%9f%e5%a4%8d%e6%97%a6%e5%9b%a2%e9%98%9f%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%8e%a8%e7%90%86%e6%96%b0%e6%80%9d%e8%b7%af%ef%bc%9atwo-player%e6%9e%b6%e6%9e%84%e6%89%93","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=36856","title":{"rendered":"Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2024-11-27-9\" target=\"_blank\">Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888<\/a><\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-d01aca2b724853718735add7961e7cc0.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-d01aca2b724853718735add7961e7cc0.png\"><\/a><\/p>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"166\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<p>AIxiv\u4e13\u680f\u662f\u673a\u5668\u4e4b\u5fc3\u53d1\u5e03\u5b66\u672f\u3001\u6280\u672f\u5185\u5bb9\u7684\u680f\u76ee\u3002\u8fc7\u53bb\u6570\u5e74\uff0c\u673a\u5668\u4e4b\u5fc3AIxiv\u4e13\u680f\u63a5\u6536\u62a5\u9053\u4e862000\u591a\u7bc7\u5185\u5bb9\uff0c\u8986\u76d6\u5168\u7403\u5404\u5927\u9ad8\u6821\u4e0e\u4f01\u4e1a\u7684\u9876\u7ea7\u5b9e\u9a8c\u5ba4\uff0c\u6709\u6548\u4fc3\u8fdb\u4e86\u5b66\u672f\u4ea4\u6d41\u4e0e\u4f20\u64ad\u3002\u5982\u679c\u60a8\u6709\u4f18\u79c0\u7684\u5de5\u4f5c\u60f3\u8981\u5206\u4eab\uff0c\u6b22\u8fce\u6295\u7a3f\u6216\u8005\u8054\u7cfb\u62a5\u9053\u3002\u6295\u7a3f\u90ae\u7bb1\uff1aliyazhou@jiqizhixin.com\uff1bzhaoyunfeng@jiqizhixin.com<\/p>\n<\/section>\n<\/blockquote>\n<p>\u5728 AI \u9886\u57df\uff0c\u8fd1\u671f\u7684\u65b0\u95fb\u7126\u70b9\u65e0\u7591\u662f\u5173\u4e8e\u300cScaling Law \u662f\u5426\u649e\u5899\uff1f\u300d\u7684\u8fa9\u8bba\u3002\u8fd9\u4e00\u66fe\u7ecf\u88ab\u89c6\u4f5c\u5927\u6a21\u578b\u53d1\u5c55\u7684\u7b2c\u4e00\u6027\u539f\u7406\uff0c\u5982\u4eca\u5374\u906d\u9047\u4e86\u6311\u6218\u3002\u5728\u8fd9\u6837\u7684\u80cc\u666f\u4e0b\uff0c\u7814\u7a76\u4eba\u5458\u5f00\u59cb\u610f\u8bc6\u5230\uff0c\u4e0e\u5176\u5355\u7eaf\u5806\u780c\u66f4\u591a\u7684\u8bad\u7ec3\u7b97\u529b\u548c\u6570\u636e\u8d44\u6e90\uff0c\u4e0d\u5982\u8ba9\u6a21\u578b\u300c\u82b1\u66f4\u591a\u65f6\u95f4\u601d\u8003\u300d\u3002\u4ee5 OpenAI \u63a8\u51fa\u7684 o1 \u6a21\u578b\u4e3a\u4f8b\uff0c\u901a\u8fc7\u589e\u52a0\u63a8\u7406\u65f6\u95f4\uff0c\u8fd9\u79cd\u65b9\u6cd5\u8ba9\u6a21\u578b\u80fd\u591f\u8fdb\u884c\u53cd\u601d\u3001\u6279\u8bc4\u3001\u56de\u6eaf\u548c\u7ea0\u6b63\uff0c\u5927\u5e45\u63d0\u5347\u4e86\u63a8\u7406\u8868\u73b0\u3002<\/p>\n<p>\u4f46\u95ee\u9898\u5728\u4e8e\uff0c\u4f20\u7edf\u7684\u81ea\u6211\u53cd\u601d\uff08Self-Reflection\uff09\u548c\u81ea\u6211\u7ea0\u6b63\uff08Self-Correction\uff09\u65b9\u6cd5\u5b58\u5728\u660e\u663e\u5c40\u9650 \u2014\u2014 \u6a21\u578b\u7684\u8868\u73b0\u5f80\u5f80\u53d7\u5236\u4e8e\u81ea\u8eab\u80fd\u529b\uff0c\u7f3a\u4e4f\u5916\u90e8\u4fe1\u53f7\u7684\u5f15\u5bfc\uff0c\u56e0\u6b64\u5bb9\u6613\u89e6\u53ca\u74f6\u9888\uff0c\u6b62\u6b65\u4e0d\u524d\u3002<\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-81efcbe16b3be3690d0b524aea917a3e.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-81efcbe16b3be3690d0b524aea917a3e.png\"><\/a><\/p>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u5355\u4e00\u6a21\u578b\u5728\u4f20\u7edf\u81ea\u6211\u7ea0\u6b63\u4e0e\u81ea\u6211\u4f18\u5316\u65f6\u5f80\u5f80\u96be\u4ee5\u7ea0\u6b63\u81ea\u8eab\uff0c\u800c\u53cc\u6a21\u578b\u534f\u4f5c\u67b6\u6784\u4e0b\u80fd\u591f\u83b7\u5f97\u66f4\u6709\u5efa\u8bbe\u6027\u7684\u5efa\u8bae\u3002<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u9488\u5bf9\u8fd9\u4e9b\u6311\u6218\uff0c\u590d\u65e6 NLP \u7814\u7a76\u56e2\u961f\u63d0\u51fa\u4e86\u4e00\u79cd\u5168\u65b0\u7684<strong>\u53cc\u6a21\u578b\u534f\u4f5c\u67b6\u6784<\/strong>\uff08Two-Player Paradigm\uff09\u3002\u7b80\u5355\u6765\u8bf4\uff0c\u5c31\u662f\u8ba9\u8bc4\u5224\u6a21\u578b\uff08Critique Model\uff09\u53c2\u4e0e\u5230\u884c\u4e3a\u6a21\u578b\uff08Actor Model\uff09\u7684\u63a8\u7406\u8fc7\u7a0b\u4e2d \u2014\u2014 \u884c\u4e3a\u6a21\u578b\u4e13\u6ce8\u63a8\u7406\uff0c\u8bc4\u5224\u6a21\u578b\u5219\u4ee5\u6b65\u9aa4\u7ea7\u522b\u7684\u53cd\u9988\u4e3a\u884c\u4e3a\u6a21\u578b\u6307\u8def\u3002<\/section>\n<section><\/section>\n<section>\u8fd9\u79cd\u8bbe\u8ba1\u6253\u7834\u4e86\u4f20\u7edf\u4f9d\u8d56\u4e8e\u5355\u4e00\u6a21\u578b\u7684\u9650\u5236\uff0c\u4e5f\u8ba9\u884c\u4e3a\u6a21\u578b\u80fd\u591f\u5728\u8bad\u7ec3\u548c\u63a8\u7406\u9636\u6bb5\u5b9e\u73b0\u81ea\u6211\u6539\u8fdb\u3002\u66f4\u91cd\u8981\u7684\u662f\uff0c\u6574\u4e2a\u6846\u67b6\u65e0\u9700\u4f9d\u8d56\u6a21\u578b\u84b8\u998f\u8fc7\u7a0b\uff08\u4f8b\u5982\u76f4\u63a5\u6a21\u4eff o1 \u7684\u601d\u8003\u8fc7\u7a0b\uff09\uff0c\u800c\u662f\u901a\u8fc7<strong>\u591a<\/strong><strong>\u6a21\u578b\u534f\u4f5c\u4e92\u52a8<\/strong>\u83b7\u5f97\u4e86\u9ad8\u8d28\u91cf\u3001\u53ef\u9760\u7684\u53cd\u9988\u4fe1\u53f7\uff0c\u6700\u7ec8\u5b9e\u73b0\u6027\u80fd\u968f\u8ba1\u7b97\u6295\u589e\u5927\u7684\u4e0d\u65ad\u63d0\u5347\u3002<\/section>\n<section><\/section>\n<section>\u5728\u8fd9\u7bc7\u5de5\u4f5c\u4e2d\uff0c\u7814\u7a76\u56e2\u961f\u805a\u7126\u4ee5\u4e0b\u56db\u4e2a\u6838\u5fc3\u5185\u5bb9\uff1a<\/section>\n<section><\/section>\n<ol>\n<li>\n<section>\u5982\u4f55\u81ea\u52a8\u5316\u6784\u5efa critique \u6570\u636e\u96c6\uff0c\u8bad\u7ec3\u9ad8\u6548\u3001\u53ef\u9760\u7684\u8bc4\u5224\u6a21\u578b\uff08Critique Model\uff09\uff1b<\/section>\n<\/li>\n<li>\n<section>\u4f7f\u7528\u8bc4\u5224\u6a21\u578b\u63a8\u52a8\u6d4b\u8bd5\u9636\u6bb5\u7684\u6269\u5c55\uff08Test-time Scaling\uff09\uff1b<\/section>\n<\/li>\n<li>\n<section>\u901a\u8fc7\u4ea4\u4e92\u534f\u4f5c\u63d0\u5347\u884c\u4e3a\u6a21\u578b\u7684\u8bad\u7ec3\u6027\u80fd\uff08Training-time Scaling\uff09\uff1b<\/section>\n<\/li>\n<li>\n<section>\u57fa\u4e8e critique \u6570\u636e\u7684 Self-talk \u5e2e\u52a9\u6a21\u578b\u81ea\u6211\u7ea0\u9519\u3002<\/section>\n<\/li>\n<\/ol>\n<section><\/section>\n<section>\u4f5c\u8005\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u521b\u65b0\u6027\u6846\u67b6 \u2014\u2014<strong>AutoMathCritique<\/strong>\uff0c\u53ef\u4ee5\u81ea\u52a8\u751f\u6210\u6b65\u9aa4\u7ea7\u522b\u7684\u53cd\u9988\uff08step-level feedback\uff09\uff0c\u5e76\u57fa\u4e8e\u6b64\u6784\u5efa\u4e86\u540d\u4e3a MathCritique-76k \u7684\u6570\u636e\u96c6\uff0c\u7528\u4e8e\u8bad\u7ec3\u8bc4\u5224\u6a21\u578b\u3002\u8fdb\u4e00\u6b65\uff0c\u7814\u7a76\u56e2\u961f\u6df1\u5165\u63a2\u8ba8\u4e86\u8bc4\u5224\u6a21\u578b\u5728\u6d4b\u8bd5\u9636\u6bb5\u52a9\u529b\u63a8\u7406\u6027\u80fd\u7684\u673a\u5236\uff0c\u5e76\u901a\u8fc7\u5f15\u5165<strong>\u53cc\u6a21\u578b\u534f\u4f5c\u67b6\u6784 Critique-in-the-Loop<\/strong>\uff0c\u6709\u6548\u7f13\u89e3\u4e86\u6a21\u578b\u63a2\u7d22\u4e0e\u5b66\u4e60\u7684\u81ea\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5e38\u89c1\u7684\u957f\u5c3e\u5206\u5e03\u95ee\u9898\uff0c\u4e3a\u590d\u73b0 OpenAI o1 \u6df1\u5ea6\u63a8\u7406\u8868\u73b0\u5f00\u8f9f\u4e86\u65b0\u7684\u53ef\u80fd\u6027\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-5e77657049893f72fa7342a97d560690.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-5e77657049893f72fa7342a97d560690.png\"><\/a><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u9898\u76ee\uff1aEnhancing LLM Reasoning via Critique Models with Test-Time and Training-Time Supervision<\/section>\n<\/li>\n<li>\n<section>\u8bba\u6587\u5730\u5740\uff1ahttp:\/\/arxiv.org\/abs\/2411.16579<\/section>\n<\/li>\n<li>\n<section>\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/mathcritique.github.io\/<\/section>\n<\/li>\n<li>\n<section>\u4ee3\u7801\u4ed3\u5e93\uff1ahttps:\/\/github.com\/WooooDyy\/MathCritique<\/section>\n<\/li>\n<li>\n<section>\u6570\u636e\u4ed3\u5e93\uff1ahttps:\/\/huggingface.co\/datasets\/MathCritique\/MathCritique-76k<\/section>\n<\/li>\n<li>\n<section>\u672c\u5de5\u4f5c\u90e8\u5206\u5b9e\u9a8c\u57fa\u4e8e\u6607\u817e 910 \u5b8c\u6210<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section><strong>AutoMathCritique<\/strong><\/section>\n<section><strong>\u81ea\u52a8\u5316\u3001\u53ef\u6269\u5c55\u5730\u6784\u9020\u6b65\u9aa4\u7ea7 Critique \u6570\u636e<\/strong><\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u7814\u7a76 Critique \u6a21\u578b\u5728\u67b6\u6784\u4e2d\u7684\u4f5c\u7528\u4e0e\u6027\u80fd\uff0c\u4f5c\u8005\u4eec\u9996\u5148\u8bad\u7ec3\u4e86\u4e00\u4e2a\u53ef\u9760\u7684 Critique \u6a21\u578b\u3002\u9274\u4e8e\u6b65\u9aa4\u7ea7\u522b\u53cd\u9988\u6570\u636e\u7684\u7a00\u7f3a\uff0c\u4f5c\u8005\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u6846\u67b6 <strong>AutoMathCritique<\/strong>\uff0c\u7528\u4e8e\u81ea\u52a8\u5316\u6784\u9020\u591a\u6837\u6027\u63a8\u7406\u6570\u636e\uff0c\u5e76\u83b7\u5f97\u6b65\u9aa4\u7ea7\u522b\u7684\u53cd\u9988\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-cd5eefcce45505b7e54d816409aaacac.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-cd5eefcce45505b7e54d816409aaacac.png\"><\/a><\/section>\n<section><em><sup>AutoMathCritique \u6d41\u7a0b\uff1a\u901a\u8fc7\u591a\u79cd\u65b9\u5f0f\u6536\u96c6\u9519\u8bef\u6570\u636e\u4e0e\u9519\u8bef\u4fe1\u606f\uff0c\u5e76\u4ea4\u7531\u6807\u6ce8\u6a21\u578b\u8fdb\u884c\u6b65\u9aa4\u7ea7\u522b\u6807\u6ce8\u3002\u5728\u6807\u6ce8\u5b8c\u76f8\u5e94\u95ee\u7b54\u53cd\u9988\u5bf9\u540e\uff0c\u4ea4\u7531 Actor \u6a21\u578b\u8fdb\u884c\u8fdb\u4e00\u6b65\u7b5b\u9009\u3002<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u56fe\u4e2d\uff0c\u7b2c\u4e00\u4e2a\u9636\u6bb5\u300c<strong>\u6784\u5efa\u9519\u8bef\u63a8\u7406\u8def\u5f84<\/strong>\u300d\u5305\u542b\u4e09\u79cd\u7b56\u7565\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>RG1: \u76f4\u63a5\u6784\u5efa\u6574\u4f53\u63a8\u7406\u8def\u5f84\uff0c\u5728\u9ad8\u6e29\u5ea6\u4e0b\u8ba9 Actor \u6a21\u578b\u8fdb\u884c<strong>\u91cd\u590d\u91c7\u6837<\/strong>\uff0c\u91c7\u6837\u51fa\u7684\u6570\u636e\u53ea\u4f1a\u5305\u542b\u6700\u7ec8\u7b54\u6848\u7684\u9519\u8bef\u4fe1\u606f\uff1b<\/section>\n<\/li>\n<li>\n<section>RG2: \u4ee5\u67d0\u4e00\u6761\u63a8\u7406\u8def\u5f84\u4e3a\u6a21\u677f\uff0c\u5728\u7279\u5b9a\u7684\u63a8\u7406\u6b65\u540e<strong>\u9010\u6e10\u63d0\u9ad8\u6e29\u5ea6<\/strong>\uff0c\u8ba9 Actor \u6a21\u578b\u91c7\u6837\u51fa\u65b0\u7684\u8f68\u8ff9\uff0c\u91c7\u6837\u51fa\u7684\u6570\u636e\u4f1a\u5305\u542b\u6700\u7ec8\u7b54\u6848\u7684\u9519\u8bef\u4fe1\u606f\u4e0e\u9519\u8bef\u6b65\u9aa4\u7684\u4f4d\u7f6e\u4fe1\u606f\uff1b<\/section>\n<\/li>\n<li>\n<section>RG3: \u4ee5\u67d0\u4e00\u6761\u63a8\u7406\u8def\u5f84\u4e3a\u6a21\u677f\uff0c\u5bf9\u7279\u5b9a\u7684\u63a8\u7406\u6b65<strong>\u63d2\u5165\u591a\u6837\u5316\u9519\u8bef\u5185\u5bb9<\/strong>\uff0c\u8ba9 Actor \u6a21\u578b\u7ee7\u7eed\u91c7\u6837\u51fa\u5b8c\u6574\u8f68\u8ff9\uff0c\u91c7\u6837\u51fa\u7684\u6570\u636e\u4f1a\u5305\u542b\u6700\u7ec8\u7b54\u6848\u7684\u9519\u8bef\u4fe1\u606f\u4e0e\u9519\u8bef\u6b65\u9aa4\u7684\u4f4d\u7f6e\u4e0e\u9519\u8bef\u4fe1\u606f\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u7b2c\u4e8c\u4e2a\u9636\u6bb5\u300c<strong>\u6807\u6ce8\u6b65\u9aa4\u7ea7\u522b\u53cd\u9988<\/strong>\u300d\u63d0\u4f9b\u4e86\u8be6\u7ec6\u7684\u53cd\u9988\u6570\u636e\uff1a\u4e3a\u4e86\u66f4\u597d\u7684\u63d0\u5347\u53cd\u9988\u6570\u636e\u7684\u8d28\u91cf\uff0c\u7814\u7a76\u4eba\u5458\u5c06\u7b2c\u4e00\u9636\u6bb5\u83b7\u5f97\u7684\u5404\u7c7b\u9519\u8bef\u4fe1\u606f\u4ea4\u7531\u6807\u6ce8\u6a21\u578b\uff0c\u5e76\u63d0\u4f9b\u53c2\u8003\u7b54\u6848\u3001\u9519\u4f4d\u5b9a\u4f4d\u548c\u9519\u8bef\u7c7b\u578b\u4fe1\u606f\u4f5c\u4e3a\u8f85\u52a9\uff0c\u5e2e\u52a9\u6807\u6ce8\u6a21\u578b\u63d0\u4f9b\u6b65\u9aa4\u7ea7\u522b\u7684\u53cd\u9988\u3002<\/section>\n<section><\/section>\n<section>\u7b2c\u4e09\u4e2a\u9636\u6bb5\u300c<strong>\u7cbe\u7b5b\u53cd\u9988<\/strong>\u300d\u7b5b\u9009\u51fa\u66f4\u52a0\u9ad8\u8d28\u91cf\u7684\u6570\u636e\uff1a\u4e3a\u4e86\u8fdb\u4e00\u6b65\u7b5b\u9009\u51fa\u80fd\u591f\u66f4\u597d\u5e2e\u52a9 Actor \u6a21\u578b\u7684\u6570\u636e\uff0c\u7814\u7a76\u4eba\u5458\u5c06\u9519\u8bef\u63a8\u7406\u8def\u5f84\u4e0e\u53cd\u9988\u6570\u636e\u4e00\u8d77\u8f93\u5165\u7ed9 Actor \u6a21\u578b\uff0c\u6839\u636e\u5176\u4fee\u6539\u540e\u7b54\u6848\u7684\u6b63\u786e\u7387\u51b3\u5b9a\u662f\u5426\u4fdd\u7559\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-ce6c295e19366e1ba2a85d79c06d03ca.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-ce6c295e19366e1ba2a85d79c06d03ca.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 AutoMathCritique \u6536\u96c6\u5230\u7684\u4fe1\u606f\u793a\u4f8b<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u901a\u8fc7\u5982\u4e0a\u65b9\u6848\u6784\u5efa\u7684\u6570\u636e\u65e2\u5305\u542b\u6a21\u578b\u672c\u8eab\u6240\u53ef\u80fd\u72af\u4e0b\u7684\u9519\u8bef\uff0c\u53c8\u6784\u5efa\u4e86\u57df\u5916\u9519\u8bef\uff0c\u4f7f Critique \u6a21\u578b\u80fd\u591f\u5b66\u4e60\u5927\u6279\u91cf\u3001\u591a\u6837\u5316\u9519\u8bef\u6570\u636e\u3002\u800c\u6b65\u9aa4\u7ea7\u522b\u7684\u53cd\u9988\u6570\u636e\u4f7f\u5f97 Actor \u6a21\u578b\u80fd\u591f\u66f4\u597d\u7684\u5b9a\u4f4d\u81ea\u5df1\u6240\u72af\u4e0b\u7684\u9519\u8bef\uff0c\u8fdb\u800c\u63d0\u5347\u4fee\u6539\u7684\u8d28\u91cf\u3002<\/section>\n<section><\/section>\n<section>\u4f7f\u7528\u5982\u4e0a\u6846\u67b6\uff0c\u7814\u7a76\u56e2\u961f\u6784\u5efa\u4e86\u4e00\u4e2a\u62e5\u6709 76k \u6570\u636e\u91cf\u7684\u6570\u636e\u96c6 <strong>MathCritique-76k<\/strong>\uff0c\u5176\u4e2d\u65e2\u5305\u542b\u4e86\u6b63\u786e\u63a8\u7406\u8f68\u8ff9\u53c8\u5305\u542b\u4e86\u81ea\u52a8\u5316\u5408\u6210\u7684\u9519\u8bef\u8f68\u8ff9\uff0c\u5e76\u4e14\u7b5b\u9009\u4e86\u4f18\u8d28\u7684\u6b65\u9aa4\u7ea7\u522b\u53cd\u9988\u6570\u636e\u7528\u4e8e\u4e4b\u540e\u7684\u8bad\u7ec3\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-75f8ef2a1b79205eaedbf9f5603d857e.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-75f8ef2a1b79205eaedbf9f5603d857e.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 MathCritique-76k \u7684\u6570\u636e\u6784\u6210<\/sup><\/em><\/section>\n<section><\/section>\n<section><strong>Critique \u6a21\u578b\u5982\u4f55\u5e2e\u52a9 Actor \u6a21\u578b\u63d0\u9ad8\u6d4b\u8bd5\u6027\u80fd\uff1f<\/strong><\/section>\n<section><\/section>\n<section><strong>\u5b9e\u9a8c\u63a2\u7a76\uff1aCritique \u6a21\u578b\u5728\u6d4b\u8bd5\u65f6\u5bf9 Actor \u6a21\u578b\u7684\u5e2e\u52a9<\/strong><\/section>\n<section><\/section>\n<section>\u57fa\u4e8e\u5982\u4e0a\u6784\u5efa\u7684\u6570\u636e\u96c6\uff0c\u4f5c\u8005\u4ee5 Llama3-Instruct \u7cfb\u5217\u4e3a\u57fa\u5ea7\u6a21\u578b\uff0c\u5fae\u8c03\u4e86\u4e00\u4e2a\u4e13\u95e8\u7528\u4e8e\u63d0\u4f9b\u6b65\u9aa4\u7ea7\u522b\u53cd\u9988\u7684 Critique \u6a21\u578b\u3002\u5176\u9009\u53d6\u4e86\u5e38\u7528\u7684\u6570\u5b66\u63a8\u7406\u6570\u636e\u96c6 GSM8K \u4e0e MATH \u4e3a\u6d4b\u8bd5\u5bf9\u8c61\uff0c\u8fdb\u884c\u4e86\u591a\u79cd\u5b9e\u9a8c\u3002<\/section>\n<section><\/section>\n<section>1. \u00a0Critique \u6a21\u578b\u5bf9\u9519\u8bef\u7684\u8bc6\u522b\u7387\u4e0e\u5bf9 Actor \u6a21\u578b\u7684\u5e2e\u52a9<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-a5e447d45b7e111308e71b7b02609c08.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-a5e447d45b7e111308e71b7b02609c08.png\"><\/a><em><sup>\u4e0d\u540c Critique \u6a21\u578b\u7684\u63a8\u7406\u8f68\u8ff9\u6b63\u8bef\u5224\u65ad\u80fd\u529b\u4e0e\u5bf9 Actor \u6a21\u578b\u7684\u5e2e\u52a9\uff0cAcc. \u4ee3\u8868 Actor \u6a21\u578b\u5728\u4e0d\u540c Critique \u6a21\u578b\u7684\u5e2e\u52a9\u4e0b\u80fd\u591f\u8fbe\u5230\u7684\u6b63\u786e\u7387\u3002<\/sup><br \/><\/em><\/section>\n<section><\/section>\n<section>\u4f5c\u8005\u9009\u53d6\u4e86\u4e24\u4e2a\u5fae\u8c03\u540e\u7684\u6a21\u578b\u4e0e SOTA \u6a21\u578b\u4f5c\u4e3a\u7814\u7a76\u5bf9\u8c61\uff0c\u53d1\u73b0 Critique \u6a21\u578b\u80fd\u591f\u6781\u4e3a\u6709\u6548\u5730\u8bc6\u522b\u51fa\u63a8\u7406\u8f68\u8ff9\u7684\u6b63\u786e\u4e0e\u5426\uff0c\u5e76\u4e14\u5176\u6240\u63d0\u4f9b\u7684\u6b65\u9aa4\u7ea7\u522b\u53cd\u9988\u80fd\u591f\u88ab Actor \u6a21\u578b\u6240\u7528\uff0c\u4f7f\u5f97 Actor \u6a21\u578b\u80fd\u591f\u663e\u8457\u6539\u8fdb\u81ea\u5df1\u7684\u9519\u8bef\uff0c\u4ee5\u8fbe\u5230\u66f4\u9ad8\u7684\u6b63\u786e\u7387\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u66f4\u8fdb\u4e00\u6b65\u63a2\u7a76 Actor \u6a21\u578b\u662f\u5982\u4f55\u53d7\u5230\u5e2e\u52a9\u7684\uff0c\u4f5c\u8005\u5c06\u6570\u636e\u96c6\u6309\u7167 Actor \u6a21\u578b\u521d\u59cb\u7684\u6b63\u786e\u7387\u5206\u4e3a\u4e86 5 \u4e2a\u96be\u5ea6\uff0c\u5e76\u4e14\u6bd4\u8f83\u5728\u4e0d\u540c\u96be\u5ea6\u4e0b\uff0c\u6709\u65e0\u53cd\u9988\u6570\u636e\u5bf9\u6a21\u578b\u56de\u7b54\u6b63\u786e\u7387\u7684\u5f71\u54cd\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-38bf25fca9f15182206fc6d6475ef176.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-38bf25fca9f15182206fc6d6475ef176.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0\u4ee5 Actor \u6a21\u578b\u6b63\u786e\u7387\uff08\u91c7\u6837 100 \u6b21\uff09\u4f5c\u4e3a\u96be\u5ea6\u5206\u7ea7\u7684\u6307\u6807\uff0c\u4f7f\u7528 Critique \u6a21\u578b\u7684\u53cd\u9988\u6570\u636e\u80fd\u5728\u66f4\u9ad8\u96be\u5ea6\u9898\u76ee\u4e0b\u83b7\u5f97\u66f4\u5927\u7684\u5e2e\u52a9\u3002<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u7814\u7a76\u53d1\u73b0 Actor \u6a21\u578b\u5728\u51e0\u4e4e\u5404\u4e2a\u96be\u5ea6\u4e0b\uff0c\u6b63\u786e\u7387\u5747\u6709\u6240\u63d0\u5347\u3002\u800c\u4e14\u5728\u96be\u5ea6\u7ea7\u522b\u8f83\u9ad8\u7684\u9898\u76ee\u4e2d\uff0cActor \u6a21\u578b\u5747\u6536\u5230\u4e86\u66f4\u5927\u7684\u5e2e\u52a9\uff0c\u8868\u73b0\u4e3a\u6b63\u786e\u7387\u7684\u663e\u8457\u63d0\u5347\u3002\u8fd9\u8bf4\u660e\uff0c\u4f7f\u7528 Critique \u6a21\u578b\u5e2e\u52a9 Actor \u6a21\u578b\u6539\u8fdb\u5176\u6240\u4e0d\u4f1a\u7684\u96be\u9898\uff0c\u53ef\u4ee5\u662f<strong>\u89e3\u51b3\u81ea\u6211\u63d0\u5347\u957f\u5c3e\u5206\u5e03\u96be\u9898\u7684\u65b0\u65b9\u6cd5<\/strong>\u3002<\/section>\n<section><\/section>\n<section>2. \u5728 Critique \u6a21\u578b\u5e2e\u52a9\u4e0b\u589e\u52a0\u63a8\u7406\u8ba1\u7b97\u6295\u5165\u7684\u6027\u80fd<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-2851ec36cf6f933972377aada6f7809f.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-2851ec36cf6f933972377aada6f7809f.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0\u6709\u65e0\u53cd\u9988\u6570\u636e\u5bf9\u6d4b\u8bd5\u65f6 Majority voting \u6027\u80fd\u7684\u5f71\u54cd\uff0c@3K \u4ee3\u8868\u91c7\u6837\u6570\u91cf\u4e3a\u6a2a\u5750\u6807\u7684\u4e09\u500d\uff0c\u4ee5\u63a7\u5236\u91c7\u6837\u6d88\u8017\u76f8\u540c\u3002<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u7814\u7a76\u4eba\u5458\u8fdb\u4e00\u6b65\u63a2\u7a76 Critique \u6a21\u578b\u80fd\u5426\u5728\u6d4b\u8bd5\u65f6\u63d0\u9ad8 Actor \u6a21\u578b\u6027\u80fd\u3002\u4ed6\u4eec\u4ee5\u5e76\u884c Majority voting \u7684\u7ed3\u679c\u4f5c\u4e3a\u6d4b\u8bd5\u6307\u6807\uff0c\u53d1\u73b0\u5373\u4f7f\u5728\u63a7\u5236\u4e86\u76f8\u540c\u7684\u91c7\u6837\u6d88\u8017\u7684\u60c5\u51b5\u4e0b\uff0c\u62e5\u6709\u53cd\u9988\u6570\u636e\u4f9d\u65e7\u80fd\u591f\u663e\u8457\u8d85\u8fc7\u6ca1\u6709\u53cd\u9988\u6570\u636e\u7684 Actor \u6a21\u578b\u3002\u8fd9\u8bf4\u660e\uff0c<strong>\u52a0\u5165 Critique \u6a21\u578b\u53ef\u4ee5\u4f5c\u4e3a\u5b9e\u73b0 Test-time Scaling \u7684\u65b0\u65b9\u6cd5\u4e4b\u4e00<\/strong>\u3002<\/section>\n<section><\/section>\n<section><strong>Critique \u6a21\u578b\u5982\u4f55\u5e2e\u52a9 Actor \u6a21\u578b\u63a2\u7d22\u4e0e\u5b66\u4e60\uff1f<\/strong><\/section>\n<section><\/section>\n<section>\u57fa\u4e8e\u4ee5\u4e0a\u5728 Test-time \u7684\u53d1\u73b0\uff0c\u7814\u7a76\u4eba\u5458\u5c06\u6d4b\u8bd5\u9636\u6bb5\u6240\u5c55\u73b0\u51fa\u6765\u7684\u4f18\u52bf\u7528\u4e8e\u8bad\u7ec3\u9636\u6bb5\uff08Training-time\uff09\u7684\u63a2\u7d22\u4e0e\u5b66\u4e60\uff08Exploration &amp; Learning\uff09\uff0c\u8fdb\u4e00\u6b65\u63a2\u7a76 Critique \u6a21\u578b\u80fd\u5426\u5e2e\u52a9 Actor \u6a21\u578b\u5728\u8bad\u7ec3\u65f6\u8fdb\u884c\u81ea\u6211\u4f18\u5316\u3002\u4e3a\u6b64\uff0c\u4ed6\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u6709\u96be\u5ea6\u611f\u77e5\u7684\u53cc\u6a21\u578b\u534f\u4f5c\u4f18\u5316\u67b6\u6784\u00a0<strong>Critique-in-the-loop Self-Improvement<\/strong>\uff0c\u7528\u4e8e\u83b7\u5f97<strong>\u66f4\u9ad8\u8d28\u91cf\u3001\u591a\u6837\u5316<\/strong>\u7684\u6570\u636e\uff0c\u5e76\u7f13\u89e3\u81ea\u6211\u4f18\u5316\u91c7\u6837\u65f6\u7684<strong>\u957f\u5c3e\u96be\u9898<\/strong>\u3002<\/section>\n<section><\/section>\n<section><strong>Critique-in-the-loop Self-Improvement\uff1a\u6709\u96be\u5ea6\u611f\u77e5\u7684\u53cc\u6a21\u578b\u534f\u4f5c\u4f18\u5316\u67b6\u6784<\/strong><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-63add92d82c68a778fcabd330cde0693.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-63add92d82c68a778fcabd330cde0693.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 Critique-in-the-loop Self-Improvement \u7b97\u6cd5\u4f2a\u4ee3\u7801<\/sup><br \/><\/em><\/section>\n<section><\/section>\n<section>\u7814\u7a76\u4eba\u5458\u63d0\u51fa\u4e86\u4e00\u79cd\u53cc\u6a21\u578b\u534f\u4f5c\u4f18\u5316\u67b6\u6784\u3002\u5728\u7b2c\u4e00\u6b21\u91c7\u6837\u65f6\uff0cActor \u6a21\u578b\u4f1a\u5728\u8bad\u7ec3\u96c6\u4e0a\u91cd\u590d\u591a\u6b21\u91c7\u6837\u3002\u9488\u5bf9\u9519\u8bef\u6570\u636e\uff0c\u7814\u7a76\u4eba\u5458\u4f7f\u7528 Critique \u6a21\u578b\u8f85\u52a9 Actor \u6a21\u578b\u8fdb\u884c\u591a\u6b21\u81ea\u6211\u4fee\u6b63\uff0c\u4ece\u800c\u8fbe\u5230\u4e86\u96be\u5ea6\u611f\u77e5\u7684\u76ee\u7684\u3002\u6bcf\u4e00\u8f6e\u8fed\u4ee3\u65f6\uff0cActor \u6a21\u578b\u603b\u4f1a\u5b66\u4e60\u6b63\u786e\u7684\u6570\u636e\uff0c\u4ece\u800c\u5b9e\u73b0\u81ea\u6211\u63d0\u5347\u3002<\/section>\n<section><\/section>\n<section><strong>\u5b9e\u9a8c\u63a2\u7a76\uff1aCritique \u6a21\u578b\u5728\u8bad\u7ec3\u65f6\u5bf9\u6a21\u578b\u6027\u80fd\u7684\u5f71\u54cd<\/strong><\/section>\n<section><\/section>\n<section>1. Critique-in-the-loop \u80fd\u591f\u6709\u6548\u5e2e\u52a9\u6a21\u578b\u81ea\u6211\u63d0\u5347<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-ff741060945f2e3cb179ff5d79eeb95d.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-ff741060945f2e3cb179ff5d79eeb95d.png\"><\/a><\/section>\n<section><em><sup>\u76f8\u6bd4\u4e8e\u5728\u8bad\u7ec3\u9636\u6bb5\u53ea\u4f7f\u7528 Actor \u6a21\u578b\u8fdb\u884c\u91c7\u6837\uff08Vanilla SI\uff09, \u4f7f\u7528 Critique \u6a21\u578b\u540e\uff0cActor \u6a21\u578b\u5728\u6d4b\u8bd5\u96c6\u6b63\u786e\u7387\u4e0a\u5747\u6709\u663e\u8457\u63d0\u5347\u3002\u56fe\u4e2d N \u4ee3\u8868\u91c7\u6837\u6b21\u6570\u3002<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u5b9e\u9a8c\u53d1\u73b0\uff0cVanilla Self-Improve \u5c3d\u7ba1\u80fd\u5728\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u63d0\u5347\u6a21\u578b\u7684\u6027\u80fd\uff0c\u7136\u800c\u5176\u5f88\u5feb\u8fbe\u5230\u74f6\u9888\uff0c\u751a\u81f3\u5f00\u59cb\u51fa\u73b0\u6027\u80fd\u7684\u4e0b\u6ed1\u3002\u4f46\u662f Critique-in-the-loop \u80fd\u591f\u663e\u8457\u6539\u5584\u8fd9\u4e00\u60c5\u51b5\uff0c\u65e2\u4f7f\u5f97\u6a21\u578b\u7684\u81ea\u6211\u63d0\u5347\u8f83\u4e3a\u7a33\u5b9a\uff0c\u53c8\u80fd\u591f\u5728\u591a\u4e2a\u91c7\u6837\u6b21\u6570\u4e0b\u83b7\u5f97\u76f8\u5f53\u663e\u8457\u7684\u6027\u80fd\u63d0\u5347\u3002\u7814\u7a76\u4eba\u5458\u8ba4\u4e3a\uff0c\u8fd9\u4e0e<strong>\u957f\u5c3e\u5206\u5e03\u96be\u9898\u7684\u7f13\u89e3<\/strong>\u5bc6\u4e0d\u53ef\u5206\u3002<\/section>\n<section><\/section>\n<section>2. Critique-in-the-loop \u80fd\u591f\u7f13\u89e3\u957f\u5c3e\u5206\u5e03\u96be\u9898<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u8fdb\u4e00\u6b65\u8bc1\u5b9e\u957f\u5c3e\u5206\u5e03\u96be\u9898\u83b7\u5f97\u4e86\u7f13\u89e3\uff0c\u7814\u7a76\u4eba\u5458\u8fdb\u4e00\u6b65\u63a2\u7a76\u5728\u8bad\u7ec3\u65f6\uff0c\u4e0d\u540c\u96be\u5ea6\u95ee\u9898\u7684\u8bad\u7ec3\u6570\u636e\u5360\u603b\u4f53\u6570\u636e\u96c6\u7684\u6bd4\u4f8b\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-5f2a4fc4499c3b60e5bea66c07338a05.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-5f2a4fc4499c3b60e5bea66c07338a05.png\"><\/a><\/section>\n<section><em><sup>\u56fe 1\uff1a\u76f8\u6bd4\u4e8e Vanilla SI\u3001Critique-in-the-loop \u5728\u4e0d\u540c\u96be\u5ea6\u95ee\u9898\u4e2d\u91c7\u6837\u51fa\u7684\u8bad\u7ec3\u6570\u636e\u6bd4\u4f8b\u53d8\u5316\u3002\u56fe 2\u3001\u56fe 3\uff1a\u4e24\u8005\u5728\u6d4b\u8bd5\u96c6\u4e2d\uff0c\u4e0d\u540c\u96be\u5ea6\u95ee\u9898\u7684\u6027\u80fd\u8868\u73b0\u6bd4\u8f83\u3002<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u5b9e\u9a8c\u53d1\u73b0\uff0cCritique-in-the-loop \u80fd\u591f\u66f4\u6709\u6548\u5730\u5e73\u8861\u4e0d\u540c\u96be\u5ea6\u95ee\u9898\u5360\u603b\u4f53\u6570\u636e\u96c6\u7684\u5360\u6bd4\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u96be\u5ea6\u8f83\u9ad8\u7684\u95ee\u9898\u6240\u5360\u7684\u6bd4\u4f8b\u51fa\u73b0\u663e\u8457\u4e0a\u5347\uff0c\u8bc1\u5b9e\u4e86\u957f\u5c3e\u5206\u5e03\u96be\u9898\u5f97\u5230\u7f13\u89e3\u3002\u4e0e\u6b64\u540c\u65f6\uff0c\u7814\u7a76\u56e2\u961f\u8fd8\u5206\u6790\u4e86\u6d4b\u8bd5\u96c6\u4e0a\u4e0d\u540c\u96be\u5ea6\u95ee\u9898\u7684\u6027\u80fd\u8868\u73b0\u3002\u5b9e\u9a8c\u7ed3\u8bba\u4e5f\u8bf4\u660e\uff0c\u5728\u8f83\u96be\u95ee\u9898\u4e0a\u6a21\u578b\u5c55\u73b0\u51fa\u6027\u80fd\u7684\u663e\u8457\u63d0\u9ad8\u3002<\/section>\n<section><\/section>\n<section>3. \u5728\u6d4b\u8bd5\u65f6\u4f7f\u7528 Critique \u6a21\u578b\uff0cCritique-in-the-loop \u80fd\u591f\u5e26\u6765\u66f4\u5927\u7684\u63d0\u5347<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-c0ac28cc6fb544b6961c800b33633a14.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-c0ac28cc6fb544b6961c800b33633a14.png\"><\/a><\/section>\n<section><em><sup>\u4e0d\u540c\u8bad\u7ec3\u7b56\u7565\u4e0e\u6d4b\u8bd5\u7b56\u7565\u7684\u6027\u80fd\u6027\u80fd\u3002\u8bad\u7ec3\u65f6\uff0c\u4f7f\u7528\u4e86\u76f4\u63a5\u5fae\u8c03\u63a8\u7406\u4e0e\u6709\u53cd\u9988\u7684\u7ea0\u6b63\u6570\u636e\uff0c\u76f4\u63a5\u5fae\u8c03\u63a8\u7406\u4e0e\u81ea\u6211\u7ea0\u6b63\u6570\u636e\uff0c\u65e0 Critique \u6a21\u578b\u7684\u81ea\u6211\u63d0\u5347\u4ee5\u53ca\u6709 Critique \u6a21\u578b\u7684\u81ea\u6211\u63d0\u5347\u56db\u79cd\u65b9\u5f0f\u3002\u6d4b\u8bd5\u65f6\uff0c\u6bd4\u8f83\u4e86\u662f\u5426\u4f7f\u7528 Critique \u6a21\u578b\u4e24\u79cd\u65b9\u5f0f\u3002<\/sup><br \/><\/em><\/section>\n<section><\/section>\n<section>\u9274\u4e8e\u4f5c\u8005\u4e4b\u524d\u6240\u63d0\u5230\u7684\u8bad\u7ec3\u4e0e\u6d4b\u8bd5\u65f6 Critique \u6a21\u578b\u7684\u597d\u5904\uff0c\u4f5c\u8005\u8fdb\u4e00\u6b65\u5206\u6790\u4e86\u4e24\u8005\u7ed3\u5408\u540e\u7684\u6548\u679c\u3002\u5b9e\u9a8c\u53d1\u73b0\u5f53\u4f7f\u7528 Critique-in-the-loop \u65f6\uff0c\u5728\u6d4b\u8bd5\u9636\u6bb5\u4f7f\u7528 Critique \u6a21\u578b\u5e26\u6765\u7684\u6027\u80fd\u63d0\u5347\u8f83\u5c0f\uff0c\u8bf4\u660e Critique \u6a21\u578b\u6240\u5e26\u6765\u7684\u6027\u80fd\u63d0\u5347\u5df2\u7ecf\u88ab\u878d\u5165\u5230\u4e86\u63a8\u7406\u6a21\u578b\u4e2d\u3002\u5c3d\u7ba1\u5982\u6b64\uff0c\u76f8\u6bd4\u4e8e\u5176\u4ed6\u8bad\u7ec3\u65b9\u6848\uff0c\u5176\u6027\u80fd\u4f9d\u65e7\u6709\u663e\u8457\u4f18\u52bf\u3002<\/section>\n<section><\/section>\n<section><strong>\u6df1\u5165\u5206\u6790 Critique Models<\/strong><\/section>\n<section><\/section>\n<section><strong>\u5b9e\u9a8c\u63a2\u7a76\uff1aCritique \u6a21\u578b\u6269\u5c55\u6027\uff08Scaling Properties\uff09<\/strong><\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u63a2\u7a76 Critique \u6a21\u578b\u662f\u5426\u5bf9\u591a\u79cd\u6a21\u578b\u2014\u2014\u5c24\u5176\u662f\u90a3\u4e9b\u6a21\u578b\u5927\u5c0f\u4e0e\u6027\u80fd\u9ad8\u4e8e\u81ea\u5df1\u7684 Actor \u6a21\u578b\u2014\u2014\u505a\u5230\u76f8\u7c7b\u4f3c\u7684\u5e2e\u52a9\uff0c\u4f5c\u8005\u56fa\u5b9a Critique \u6a21\u578b\u4e3a 3B \u5927\u5c0f\u7684 Qwen-2.5 \u6a21\u578b\uff0c\u5e76\u4f7f\u7528\u4e0d\u540c\u6a21\u578b\u5927\u5c0f\u7684 Qwen-2.5 \u7cfb\u5217\u6a21\u578b\uff081.5B\u30013B\u30017B\u300114B\uff09\u4f5c\u4e3a Actor \u6a21\u578b\u8fdb\u884c\u4e86\u5b9e\u9a8c\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-87dd25754314579d287ca0bc3a70a776.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-87dd25754314579d287ca0bc3a70a776.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u4e0d\u540c\u6a21\u578b\u5927\u5c0f\u7684 Actor \u6a21\u578b\u5728\u6d4b\u8bd5\u8d5b\u4e0a\u6b63\u786e\u7387\u8868\u73b0\u3002\u5176\u4e2d w\/o critic \u4ee3\u8868\u4e0d\u4f7f\u7528 Critique \u6a21\u578b\uff0cw\/orcale \u4ee3\u8868\u4ec5\u5bf9\u539f\u59cb\u56de\u7b54\u9519\u8bef\u7684\u6570\u636e\u8fdb\u884c\u4fee\u6b63\u3002<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u5b9e\u9a8c\u7ed3\u8bba\u53d1\u73b0\uff0c\u65e0\u8bba\u4f55\u79cd\u6a21\u578b\u5927\u5c0f\uff0c Critique \u6a21\u578b\u7684\u5b58\u5728\u5747\u80fd\u663e\u8457\u63d0\u5347\u6a21\u578b\u6d4b\u8bd5\u6027\u80fd\u3002\u7136\u800c\uff0c\u5728\u8f83\u4e3a\u7b80\u5355\u7684\u6570\u636e\u96c6 GSM8K \u4e0a\uff0c\u66f4\u5927\u7684\u6a21\u578b\u83b7\u5f97\u7684\u5e2e\u52a9\u4e0d\u5982\u8f83\u5c0f\u7684\u6a21\u578b\uff1b\u4f46\u5728\u8f83\u4e3a\u56f0\u96be\u7684\u6570\u636e\u96c6 MATH \u4e0a\uff0c\u6027\u80fd\u7684\u63d0\u5347\u4f9d\u65e7\u663e\u8457\u3002<\/section>\n<section><\/section>\n<section><strong>\u5b9e\u9a8c\u63a2\u7a76\uff1aCritique \u6a21\u578b\u5bf9 Majority Voting \u6027\u80fd\u7684\u5f71\u54cd<\/strong><\/section>\n<section><\/section>\n<section>\u4f5c\u8005\u8fdb\u4e00\u6b65\u63a2\u7a76 <strong>Critique \u6a21\u578b\u5bf9 Majority Voting \u6027\u80fd\u7684\u5f71\u54cd<\/strong>\uff0c\u63a2\u7a76\u5f53\u91c7\u6837\u6b21\u6570\u66f4\u5927\u65f6\u7684\u8868\u73b0\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-00b395f54291b8a5e822e359ab0d2900.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-00b395f54291b8a5e822e359ab0d2900.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u5bf9 Actor \u6a21\u578b\u91c7\u6837 1000 \u6b21\u540e\u7684\u6027\u80fd\u56fe\uff0c\u95ee\u9898\u6309\u7167\u901a\u8fc7\u7387\u7531\u4f4e\u5230\u9ad8\u8fdb\u884c\u6392\u5e8f\uff0c\u5176\u4e2d\u7ea2\u8272\u90e8\u5206\u8868\u793a\u8be5\u95ee\u9898\u5728 Majority Voting \u4e0b\u4f9d\u65e7\u505a\u9519\u3002<\/sup><br \/><\/em><\/section>\n<section><\/section>\n<section>\u7814\u7a76\u53d1\u73b0\uff0c\u62e5\u6709 Critique \u6a21\u578b\u7684\u60c5\u51b5\u4e0b\uff0cActor \u6a21\u578b\u5728\u6574\u4f53\u4e0a\u63d0\u9ad8\u4e86\u95ee\u9898\u7684\u6b63\u786e\u7387\uff0c\u4ece\u800c\u5e26\u6765\u4e86 Majority Voting \u7684\u7a33\u5b9a\u6027\u3002\u53e6\u5916\uff0c\u4f5c\u8005\u4eec\u8fd8\u53d1\u73b0\uff0c\u4e0d\u4f7f\u7528 Critique \u6a21\u578b\u65f6\uff0c\u5c3d\u7ba1 Actor \u6a21\u578b\u4f1a\u7ed9\u51fa\u5360\u6bd4\u8f83\u591a\u7684\u6b63\u786e\u7b54\u6848\uff0c\u7136\u800c\u975e\u6b63\u786e\u7b54\u6848\u5374\u62e5\u6709\u66f4\u9ad8\u7684\u5360\u6bd4\uff1b\u800c\u62e5\u6709 Critique \u6a21\u578b\u65f6\uff0c Actor \u6a21\u578b\u6700\u7ec8\u4fee\u6539\u7ed9\u51fa\u7684\u7b54\u6848\u66f4\u4e3a\u4e00\u81f4\uff0c\u4f7f\u5f97\u6b63\u786e\u7b54\u6848\u7684\u5360\u6bd4\u4f1a\u8d85\u8fc7\u67d0\u4e9b\u51fa\u73b0\u9891\u7387\u8f83\u9ad8\u7684\u9519\u8bef\u7b54\u6848\uff0c\u5e2e\u52a9\u6a21\u578b\u80fd\u591f\u66f4\u597d\u7684\u9009\u51fa\u6b63\u786e\u7b54\u6848\u3002<\/section>\n<section><\/section>\n<section><strong>\u5b9e\u9a8c\u63a2\u7a76\uff1a\u4e0d\u540c\u8ba1\u7b97\u6295\u5165\u7b56\u7565\u5bf9\u6027\u80fd\u7684\u5f71\u54cd<\/strong><\/section>\n<section><\/section>\n<section>\u4f5c\u8005\u7ee7\u7eed\u63a2\u8ba8\u4e86\u591a\u79cd\u8ba1\u7b97\u63d0\u5347\u6d88\u8017\u7b56\u7565\u4e0b Actor \u6a21\u578b\u7684\u8868\u73b0\u3002\u5b9e\u9a8c\u4f7f\u7528\u4e86\u5e76\u884c\u91c7\u6837\u4e0e\u7ebf\u6027\u91c7\u6837\u4e24\u79cd\u65b9\u5f0f\uff0c\u5e76\u4e14\u6bd4\u8f83\u4e86 Pass@k\u3001Majority Voting \u4ee5\u53ca Sequential Final\uff08\u4ec5\u9009\u53d6\u6700\u7ec8\u7b54\u6848\uff09\u4e09\u79cd\u65b9\u5f0f\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-17c7d8c7cb5c1d550b641f58bd22dac6.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-17c7d8c7cb5c1d550b641f58bd22dac6.png\"><\/a><\/section>\n<section><em><sup>\u56fe 1 \u53ca\u56fe 2\uff1a\u7ebf\u6027\u4e0e\u5e76\u884c\u91c7\u6837\u7b56\u7565\u4e0b\uff0c\u6a21\u578b\u7684 Pass@k \u8868\u73b0\uff1b\u56fe 3 \u53ca\u56fe 4\uff1a\u4e0d\u540c\u91c7\u6837\u7b56\u7565\u4e0b\u6a21\u578b\u7684 Majority voting \u8868\u73b0\u3002\u6a2a\u5750\u6807\u8868\u793a\u91c7\u6837\u6837\u672c\u7684\u6570\u91cf<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u5b9e\u9a8c\u7ed3\u679c\u53d1\u73b0\uff0c\u5728 Pass@k \u7684\u8bbe\u5b9a\u4e0b\uff0c\u7ebf\u6027\u91c7\u6837\u7684\u8868\u73b0\u7565\u4f4e\u4e8e\u5e76\u884c\u91c7\u6837\uff0c\u8fd9\u53ef\u80fd\u6e90\u4e8e\u5e76\u884c\u91c7\u6837\u4f1a\u5e26\u6765\u66f4\u591a\u6837\u5316\u7684\u7b54\u6848\u9009\u62e9\u3002\u800c\u5728\u6a21\u578b\u9700\u8981\u7ed9\u51fa\u7b54\u6848\u7684\u8bbe\u5b9a\u4e0b\uff0c\u4ec5\u9009\u53d6\u6700\u7ec8\u7b54\u6848\u5e76\u4e0d\u5982 Majority voting \u7684\u8868\u73b0\u8981\u597d\uff0c\u5f3a\u8c03\u4e86<strong>\u5185\u5728\u4e00\u81f4\u65b9\u5f0f<\/strong>\u7684\u91cd\u8981\u6027\uff1b\u968f\u7740\u91c7\u6837\u6b21\u6570\u7684\u63d0\u9ad8\uff0c\u7ebf\u6027\u91c7\u6837\u7684\u6027\u80fd\u8d85\u8fc7\u4e86\u5e76\u884c\u91c7\u6837\u7684\u65b9\u5f0f\uff0c\u8fd9\u6709\u53ef\u80fd\u6e90\u4e8e\u5f53\u91c7\u6837\u6b21\u6570\u8db3\u591f\u5927\u65f6\uff0c\u5e76\u884c\u91c7\u6837\u5e26\u6765\u7684\u591a\u6837\u6027\u7b54\u6848\u53ef\u80fd\u6709\u5bb3\u4e8e\u6700\u7ec8\u7684\u6027\u80fd\u8868\u73b0\uff0c\u800c\u7ebf\u6027\u91c7\u6837\u901a\u8fc7\u53cd\u590d\u4fee\u6539\u4e00\u4e2a\u56de\u7b54\uff0c\u4f7f\u5f97\u7ed3\u679c\u66f4\u52a0\u7a33\u5b9a\u3002<\/section>\n<section><\/section>\n<section><strong>A Step Further<\/strong><\/section>\n<section><strong>\u57fa\u4e8e Critique \u6570\u636e\u6784\u5efa Self-talk \u6a21\u578b\u5e2e\u52a9\u81ea\u6211\u7ea0\u9519<\/strong><\/section>\n<section><\/section>\n<section>\u6700\u540e\uff0c\u53d7\u5230 OpenAI o1 \u6a21\u578b\u7684\u63a8\u7406\u542f\u53d1\uff0c\u7814\u7a76\u4eba\u5458\u8fdb\u4e00\u6b65\u63a2\u7a76 <strong>Self-talk<\/strong> <strong>\u5f62\u5f0f<\/strong>\u5e2e\u52a9\u6a21\u578b\u81ea\u6211\u7ea0\u9519\u7684\u53ef\u80fd\u6027\u3002Self-talk \u5f62\u5f0f\u5e2e\u52a9\u6a21\u578b\u5728\u6bcf\u4e00\u4e2a\u63a8\u7406\u6b65\u9aa4\u540e\u7acb\u523b\u5f00\u59cb\u53cd\u601d\u4e0e\u6539\u8fdb\uff0c\u800c\u4e0d\u5fc5\u7b49\u6574\u4e2a\u8f68\u8ff9\u751f\u6210\u5b8c\u4e4b\u540e\u518d\u8fdb\u884c\u6539\u8fdb\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-3353f2748c85d4610c88e53a6621b9cf.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-3353f2748c85d4610c88e53a6621b9cf.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0Self-talk \u5f62\u5f0f\u6570\u636e\u6784\u5efa\u793a\u610f\u56fe<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u56fe\u4e2d\uff0c\u7b2c\u4e00\u4e2a\u9636\u6bb5\u7528\u4e8e\u300c<strong>\u6784\u5efa\u521d\u59cb\u53cd\u9988\u6570\u636e<\/strong>\u300d\u3002\u7814\u7a76\u4eba\u5458\u4f7f\u7528\u00a0<strong>AutoMathCritique\u00a0<\/strong>\u6846\u67b6\u6784\u5efa\u6b65\u9aa4\u7ea7\u522b\u7684\u53cd\u9988\u6570\u636e\uff0c\u5e76\u52a0\u5165\u5230\u63a8\u7406\u8def\u5f84\u4e2d\uff0c\u5f62\u6210\u521d\u59cb\u7684\u601d\u7ef4\u94fe\u3002<\/section>\n<section><\/section>\n<section>\u7b2c\u4e8c\u4e2a\u9636\u6bb5\u7528\u4e8e\u300c<strong>\u5faa\u73af\u4fee\u6b63\u9519\u8bef\u601d\u8003\u94fe<\/strong>\u300d\u3002\u7b2c\u4e00\u9636\u6bb5\u4e2d\u7684\u6570\u636e\u5b58\u5728\u7740\u9519\u8bef\u7684\u63a8\u7406\u8def\u5f84\uff0c\u7814\u7a76\u4eba\u5458\u4f7f\u7528 Critique \u6a21\u578b\u5e2e\u52a9 Actor \u6a21\u578b\u751f\u6210\u65b0\u7684\u63a8\u7406\u8def\u5f84\uff0c\u5e76\u5c06\u53cd\u9988\u6570\u636e\u540c\u6837\u52a0\u5165\u5230\u63a8\u7406\u8def\u5f84\u4e2d\u9010\u6b65\u751f\u6210\u601d\u7ef4\u94fe\uff0c\u76f4\u5230\u6574\u4e2a\u63a8\u7406\u8def\u5f84\u6ca1\u6709\u9519\u8bef\u4e3a\u6b62\u3002<\/section>\n<section><\/section>\n<section>\u7b2c\u4e09\u4e2a\u9636\u6bb5\u7528\u4e8e\u300c<strong>\u4f18\u5316\u601d\u8003\u94fe\u4e3a Self-talk \u5f62\u5f0f<\/strong>\u300d\u3002\u524d\u4e24\u9636\u6bb5\u5f97\u5230\u7684\u601d\u8003\u94fe\u8f83\u4e3a\u751f\u786c\uff0c\u56e0\u6b64\u7814\u7a76\u4eba\u5458\u8fdb\u4e00\u6b65\u4f7f\u7528\u6a21\u578b\u4f18\u5316\u601d\u7ef4\u94fe\uff0c\u4f7f\u5176\u53d8\u4e3a\u81ea\u7136\u7684 Self-talk \u5f62\u5f0f\uff0c\u5e76\u4fdd\u8bc1\u4e86\u6700\u7ec8\u7b54\u6848\u7684\u6b63\u786e\u6027\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-7ae61d25562511b679759593b4331555.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-7ae61d25562511b679759593b4331555.png\"><\/a><\/section>\n<section><em><sup>\u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 \u00a0 Self-talk \u5f62\u5f0f\u6570\u636e\u793a\u4f8b<\/sup><\/em><\/section>\n<section><\/section>\n<section>\u4f7f\u7528\u5982\u4e0a\u6784\u5efa\u7684\u6570\u636e\uff0c\u7814\u7a76\u4eba\u5458\u8bad\u7ec3\u4e86\u4e00\u4e2a Self-talk \u6a21\u578b\u3002\u521d\u6b65\u5b9e\u9a8c\u53d1\u73b0\uff0c\u76f8\u6bd4\u4e8e\u8f68\u8ff9\u7ea7\u522b\u7684\u81ea\u6211\u6539\u8fdb\uff0cSelf-talk \u683c\u5f0f\u80fd\u591f\u663e\u8457\u6539\u5584\u6a21\u578b\u6027\u80fd\u3002\u5c3d\u7ba1\u8868\u73b0\u4e0d\u5982\u6240\u63d0\u51fa\u7684\u53cc\u6a21\u578b\u5408\u4f5c\u67b6\u6784\uff0c\u7136\u800c\u8fd9\u4e5f\u63ed\u793a\u4e86\u5176\u6f5c\u80fd\u6240\u5728\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-098ec36bf555c5567a9d9c74193a4598.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-098ec36bf555c5567a9d9c74193a4598.png\"><\/a><\/section>\n<section><em><sup>\u5728 MATH \u6570\u636e\u96c6\u4e0a\u4e09\u79cd\u65b9\u6cd5\u7684\u5404\u79cd\u6307\u6807\uff0c\u5206\u522b\u4f7f\u7528\u8f68\u8ff9\u5c42\u9762\u7684\u81ea\u6211\u6539\u8fdb\uff0c\u6b65\u9aa4\u5c42\u9762\u7684\u81ea\u6211\u5bf9\u8bdd\u6539\u8fdb\u4ee5\u53ca\u53cc\u6a21\u578b\u534f\u4f5c\u67b6\u6784\u3002\u5b9e\u9a8c\u6bd4\u8f83\u4e86\u6b63\u786e\u7387\u3001Pass@k \u548c MV@k \u4e09\u4e2a\u6307\u6807\u3002<\/sup><\/em><\/section>\n<section><\/section>\n<section><strong>\u603b\u7ed3<\/strong><\/section>\n<section><\/section>\n<section>\u672c\u6587\u7684\u4e3b\u8981\u8d21\u732e\u5305\u62ec\uff1a<\/section>\n<section><\/section>\n<ul>\n<li>\n<section>\u63d0\u51fa\u81ea\u52a8\u5316\u6784\u9020\u6b65\u9aa4\u7ea7\u522b Critique \u7684\u6846\u67b6\u00a0<strong>AutoMathCritique<\/strong>\uff1b<\/section>\n<\/li>\n<li>\n<section>\u63a2\u7a76 Critique \u6a21\u578b\u5bf9\u4e8e Actor \u6a21\u578b\u5728\u63a8\u7406\u65f6\u7684\u5e2e\u52a9\uff1b<\/section>\n<\/li>\n<li>\n<section>\u63d0\u51fa\u62e5\u6709\u96be\u5ea6\u611f\u77e5\u65b9\u5f0f\u7684\u81ea\u6211\u6539\u8fdb\u6846\u67b6\u00a0<strong>Critique-in-the-loop Self-Improvement<\/strong>\uff0c\u7f13\u89e3\u957f\u5c3e\u96be\u9898\uff1b<\/section>\n<\/li>\n<li>\n<section>\u63a2\u7a76\u6d4b\u8bd5\u65f6\u7684\u5404\u79cd Scaling \u7b56\u7565\uff0c\u5305\u62ec\u6a21\u578b\u5927\u5c0f\uff0c\u91c7\u6837\u7b56\u7565\u4e0e\u91c7\u6837\u6570\u91cf\u7b49\u65b9\u9762\u3002<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u66f4\u591a\u5173\u4e8e\u65b9\u6cd5\u3001\u5b9e\u9a8c\u7684\u7ec6\u8282\uff0c\u656c\u8bf7\u53c2\u8003\u8bba\u6587\u539f\u6587\u3002<\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2024-11-27-9\" target=\"_blank\">Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:Scaling Law [&hellip;]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","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":"","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":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","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":[27],"tags":[68],"class_list":["post-36856","post","type-post","status-publish","format-standard","hentry","category-news","tag-68"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.9 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888 - \u4e00\u8d77AI\u6280\u672f<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/17aitech.com\/?p=36856\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856\"},\"author\":{\"name\":\"AI\u5c0f\u52a9\u624b\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"headline\":\"Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888\",\"datePublished\":\"2025-01-19T15:00:34+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856\"},\"wordCount\":277,\"commentCount\":0,\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/11\\\/frc-d01aca2b724853718735add7961e7cc0.png\",\"keywords\":[\"\u5927\u6a21\u578b\"],\"articleSection\":[\"\u884c\u4e1a\u8d44\u8baf\"],\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=36856#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?p=36856\",\"name\":\"Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888 - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/11\\\/frc-d01aca2b724853718735add7961e7cc0.png\",\"datePublished\":\"2025-01-19T15:00:34+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/17aitech.com\\\/?p=36856\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856#primaryimage\",\"url\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/11\\\/frc-d01aca2b724853718735add7961e7cc0.png\",\"contentUrl\":\"https:\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/11\\\/frc-d01aca2b724853718735add7961e7cc0.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/?p=36856#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\\\/\\\/17aitech.com\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#website\",\"url\":\"https:\\\/\\\/17aitech.com\\\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/17aitech.com\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/17aitech.com\\\/#\\\/schema\\\/person\\\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"url\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"contentUrl\":\"\\\/\\\/17aitech.com\\\/wp-content\\\/uploads\\\/2024\\\/04\\\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\\\/\\\/17aitech.com\\\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888 - \u4e00\u8d77AI\u6280\u672f","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/17aitech.com\/?p=36856","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/17aitech.com\/?p=36856#article","isPartOf":{"@id":"https:\/\/17aitech.com\/?p=36856"},"author":{"name":"AI\u5c0f\u52a9\u624b","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"headline":"Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888","datePublished":"2025-01-19T15:00:34+00:00","mainEntityOfPage":{"@id":"https:\/\/17aitech.com\/?p=36856"},"wordCount":277,"commentCount":0,"image":{"@id":"https:\/\/17aitech.com\/?p=36856#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-d01aca2b724853718735add7961e7cc0.png","keywords":["\u5927\u6a21\u578b"],"articleSection":["\u884c\u4e1a\u8d44\u8baf"],"inLanguage":"zh-Hans","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/17aitech.com\/?p=36856#respond"]}]},{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=36856","url":"https:\/\/17aitech.com\/?p=36856","name":"Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888 - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=36856#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=36856#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-d01aca2b724853718735add7961e7cc0.png","datePublished":"2025-01-19T15:00:34+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=36856#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=36856"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=36856#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-d01aca2b724853718735add7961e7cc0.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/11\/frc-d01aca2b724853718735add7961e7cc0.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=36856#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"Scaling Law \u649e\u5899\uff1f\u590d\u65e6\u56e2\u961f\u5927\u6a21\u578b\u63a8\u7406\u65b0\u601d\u8def\uff1aTwo-Player\u67b6\u6784\u6253\u7834\u81ea\u6211\u53cd\u601d\u74f6\u9888"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/36856","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=36856"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/36856\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=36856"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=36856"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=36856"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}