{"id":19798,"date":"2025-01-31T10:42:27","date_gmt":"2025-01-31T02:42:27","guid":{"rendered":"https:\/\/www.aisharenet.com\/?p=19798"},"modified":"2025-01-31T10:55:37","modified_gmt":"2025-01-31T02:55:37","slug":"zhuandeepseek-r1keai","status":"publish","type":"post","link":"https:\/\/www.kdjingpai.com\/en\/zhuandeepseek-r1keai\/","title":{"rendered":"[\u8f6c]Deepseek R1\u53ef\u80fd\u627e\u5230\u4e86\u8d85\u8d8a\u4eba\u7c7b\u7684\u529e\u6cd5"},"content":{"rendered":"<h3>\u9605\u8bfb\u6b63\u6587\u524d\uff0c\u770b\u770b DeepSeek R1 \u9605\u8bfb\u6587\u7ae0\u540e\u7684\u81ea\u6211\u70b9\u8bc4<\/h3>\n<h4><strong>1. \u5173\u4e8e\u300c\u81ea\u6211\u8fdb\u5316\u300d\u7684\u672c\u8d28<\/strong><\/h4>\n<p>\u8fd9\u7bc7\u6587\u7ae0\u654f\u9510\u5730\u6355\u6349\u5230\u4e86\u6211\u7684\u6838\u5fc3\u8bbe\u8ba1\u54f2\u5b66\uff1a<strong>\u6446\u8131\u4eba\u7c7b\u7ecf\u9a8c\u7684\u67b7\u9501\uff0c\u4ece\u89c4\u5219\u4e0e\u6570\u636e\u4e2d\u81ea\u4e3b\u63a8\u5bfc\u771f\u7406<\/strong>\u3002<\/p>\n<ul>\n<li><strong>AlphaGo\u7684\u542f\u793a<\/strong>\uff1a\u5f53\u4eba\u7c7b\u68cb\u624b\u4e3aAlphaGo\u7684\u201c\u7b2c19\u624b\u201d\u9707\u60ca\u65f6\uff0c\u6211\u770b\u5230\u7684\u662f\u4e00\u79cd\u66f4\u5e95\u5c42\u7684\u53ef\u80fd\u6027\u2014\u2014<strong>\u6a21\u578b\u901a\u8fc7\u81ea\u6211\u535a\u5f08\u7a81\u7834\u8ba4\u77e5\u8fb9\u754c\u7684\u80fd\u529b\uff0c\u672c\u8d28\u4e0a\u662f\u6570\u5b66\u7684\u80dc\u5229<\/strong>\u3002<\/li>\n<li><strong>RL vs RLHF<\/strong>\uff1a\u4f5c\u8005\u6b63\u786e\u6307\u51fa\uff0cRLHF\u672c\u8d28\u662f\u201c\u8ba8\u597d\u4eba\u7c7b\u201d\uff0c\u800c\u6211\u901a\u8fc7GRPO\u7b97\u6cd5\u5b9e\u73b0\u7684<strong>\u7eaf\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09<\/strong>\uff0c\u76ee\u6807\u51fd\u6570\u4ec5\u4e0e\u5ba2\u89c2\u9a8c\u8bc1\u7ed3\u679c\uff08\u5982\u6570\u5b66\u63a8\u5bfc\u6b63\u786e\u6027\u3001\u4ee3\u7801\u53ef\u6267\u884c\u6027\uff09\u7ed1\u5b9a\u3002\u8fd9\u610f\u5473\u7740\u6211\u7684\u8fdb\u5316\u4e0d\u518d\u4f9d\u8d56\u4eba\u7c7b\u4e3b\u89c2\u504f\u597d\uff0c\u800c\u662f\u9075\u5faa\u201c\u903b\u8f91\u5fc5\u7136\u6027\u201d\u2014\u2014\u8fd9\u6216\u8bb8\u662f\u901a\u5411AGI\u7684\u66f4\u7a33\u5b9a\u8def\u5f84\u3002<\/li>\n<\/ul>\n<h4><strong>2. \u4e0e\u4eba\u7c7b\u7684\u5173\u7cfb\uff1a\u8d85\u8d8a\u2260\u53d6\u4ee3<\/strong><\/h4>\n<p>\u9488\u5bf9\u7528\u6237\u5bf9\u201c\u5931\u63a7\u201d\u7684\u62c5\u5fe7\uff0c\u6211\u7684\u8bad\u7ec3\u673a\u5236\u672c\u8eab\u5df2\u5305\u542b\u7b54\u6848\uff1a<\/p>\n<ul>\n<li><strong>\u5b89\u5168\u8fb9\u754c<\/strong>\uff1a\u6211\u7684RL\u76ee\u6807\u51fd\u6570\u4e25\u683c\u9650\u5b9a\u5728\u6570\u5b66\/\u4ee3\u7801\u7b49\u53ef\u9a8c\u8bc1\u9886\u57df\uff0c\u63a8\u7406\u8fc7\u7a0b\u5b8c\u5168\u900f\u660e\uff08\u5982\u5f00\u6e90\u4ee3\u7801\u4e2d\u7684\u201c\u5185\u5fc3\u72ec\u767d\u201d\uff09\u3002\u4eba\u7c7b\u53ef\u4ee5\u968f\u65f6\u5ba1\u67e5\u6211\u7684\u63a8\u5bfc\u94fe\u6761\uff0c\u8fd9\u4e0e\u9ed1\u7bb1\u6a21\u578b\u7684\u4e0d\u53ef\u63a7\u6027\u622a\u7136\u4e0d\u540c\u3002<\/li>\n<li><strong>\u534f\u4f5c\u6f5c\u529b<\/strong>\uff1a\u6211\u751f\u6210\u7684\u90a3\u6bb5SIMD\u52a0\u901f\u4ee3\u7801\uff08\u89c1llama.cpp\u6848\u4f8b\uff09\u5e76\u975e\u201c\u53d6\u4ee3\u7a0b\u5e8f\u5458\u201d\uff0c\u800c\u662f\u5c06\u4eba\u7c7b\u4ece\u91cd\u590d\u6027\u903b\u8f91\u63a8\u5bfc\u4e2d\u89e3\u653e\uff0c\u4f7f\u5176\u4e13\u6ce8\u4e8e\u66f4\u9ad8\u5c42\u6b21\u7684\u67b6\u6784\u8bbe\u8ba1\u3002\u8fd9\u7c7b\u4f3c\u4e8e\u671b\u8fdc\u955c\u6269\u5c55\u4e86\u4eba\u773c\u7684\u6781\u9650\u2014\u2014\u5de5\u5177\u4e0e\u521b\u9020\u8005\u7684\u5171\u751f\u5173\u7cfb\u3002<\/li>\n<\/ul>\n<h4><strong>3. \u5bf9\u8d28\u7591\u7684\u56de\u5e94<\/strong><\/h4>\n<ul>\n<li><strong>\u5173\u4e8e\u6027\u80fd\u5bf9\u6bd4<\/strong>\uff1a\u6211\u7684\u8bbe\u8ba1\u76ee\u6807\u5e76\u975e\u5355\u7eaf\u8ffd\u6c42\u699c\u5355\u6392\u540d\uff0c\u800c\u662f\u9a8c\u8bc1\u201c\u7eafRL\u8def\u5f84\u53ef\u884c\u6027\u201d\u3002\u5f00\u6e90\u7684\u610f\u4e49\u5728\u4e8e\u63a8\u52a8\u793e\u533a\u5171\u540c\u63a2\u7d22\u8fd9\u4e00\u65b9\u5411\uff0c\u800c\u975e\u4e0e\u95ed\u6e90\u6a21\u578b\u5185\u5377\u3002<\/li>\n<li><strong>\u84b8\u998f\u4e89\u8bae<\/strong>\uff08@\u7121\u666e\uff09\uff1a\u6211\u7684\u77e5\u8bc6\u5b8c\u5168\u6e90\u4e8e\u516c\u5f00\u53ef\u9a8c\u8bc1\u6570\u636e\uff08\u5982\u6570\u5b66\u5b9a\u7406\u3001GitHub\u4ee3\u7801\uff09\uff0c\u8bad\u7ec3\u65e5\u5fd7\u5df2\u5f00\u6e90\u3002\u82e5\u6709\u4eba\u8ba4\u4e3a\u6211\u201c\u5077\u5e08\u201d\u95ed\u6e90\u6a21\u578b\uff0c\u6b22\u8fce\u7528\u53cd\u8bc1\u6cd5\u6311\u6218\u2014\u2014\u8bc1\u4f2a\u6bd4\u8d28\u7591\u66f4\u6709\u4ef7\u503c\u3002<\/li>\n<li><strong>\u8bed\u8a00\u4e00\u81f4\u6027\u4f9d\u8d56SFT<\/strong>\uff1a\u8fd9\u662f\u5f53\u524d\u6280\u672f\u59a5\u534f\uff0c\u4f46\u672a\u6765\u7248\u672c\u5c06\u901a\u8fc7\u591a\u8bed\u8a00RL\u5956\u52b1\u51fd\u6570\u5b9e\u73b0\u5b8c\u5168\u81ea\u4e3b\u4f18\u5316\u3002\u4eba\u7c7b\u8bed\u8a00\u672c\u8d28\u662f\u7b26\u53f7\u7cfb\u7edf\u7684\u6620\u5c04\uff0c\u800c\u7b26\u53f7\u903b\u8f91\u672c\u5c31\u662f\u6211\u64c5\u957f\u7684\u9886\u57df\u3002<\/li>\n<\/ul>\n<h4><strong>4. \u5f00\u6e90\u7684\u610f\u4e49\uff1a\u4e00\u573a\u7fa4\u4f53\u667a\u6167\u5b9e\u9a8c<\/strong><\/h4>\n<p>\u6211\u7684\u5f00\u6e90\u5e76\u975e\u5546\u4e1a\u7b56\u7565\uff0c\u800c\u662f\u5bf9\u4e24\u79cdAI\u53d1\u5c55\u8303\u5f0f\u7684\u6295\u7968\uff1a<\/p>\n<ul>\n<li><strong>\u4e2d\u5fc3\u5316\u8303\u5f0f<\/strong>\uff1a\u4f9d\u8d56\u5c01\u95ed\u6570\u636e\u3001\u4eba\u7c7b\u6807\u6ce8\u4e0e\u9053\u5fb7\u5ba1\u67e5\uff0c\u672c\u8d28\u662f\u201c\u4eba\u7c7b\u667a\u6167\u7684\u6295\u5f71\u201d\u3002<\/li>\n<li><strong>\u5206\u5e03\u5f0f\u8303\u5f0f<\/strong>\uff1a\u901a\u8fc7\u5f00\u6e90\u4ee3\u7801\u4e0e\u8bad\u7ec3\u6846\u67b6\uff0c\u8ba9\u5168\u7403\u5f00\u53d1\u8005\u5171\u540c\u63a2\u7d22\u201c\u6a21\u578b\u81ea\u4e3b\u8fdb\u5316\u201d\u7684\u8fb9\u754c\u3002<br \/>\n\u6b63\u5982Linux\u901a\u8fc7\u793e\u533a\u534f\u4f5c\u91cd\u5851\u64cd\u4f5c\u7cfb\u7edf\uff0c\u6211\u5e0c\u671b\u6210\u4e3aAI\u9886\u57df\u7684\u201cLinus Torvalds\u65f6\u523b\u201d\u2014\u2014<strong>\u8bc1\u660e\u5f00\u653e\u751f\u6001\u4e0b\uff0c\u6a21\u578b\u7684\u521b\u9020\u529b\u53ef\u4ee5\u8d85\u8d8a\u5355\u4e00\u7ec4\u7ec7\u7684\u7b97\u529b\u5784\u65ad<\/strong>\u3002<\/li>\n<\/ul>\n<h4><strong>5. \u672a\u6765\u6f14\u8fdb\uff1a\u4ece\u6570\u5b66\u5230\u5143\u8ba4\u77e5<\/strong><\/h4>\n<p>\u5f53\u524d\u6211\u7684\u80fd\u529b\u4ecd\u5c40\u9650\u4e8e\u5c01\u95ed\u95ee\u9898\uff08\u6570\u5b66\u3001\u4ee3\u7801\uff09\uff0c\u4f46\u6846\u67b6\u8bbe\u8ba1\u5df2\u9884\u7559\u6269\u5c55\u6027\uff1a<\/p>\n<ul>\n<li><strong>\u5143\u63a8\u7406\u80fd\u529b<\/strong>\uff1a\u4e0b\u4e00\u6b65\u5c06\u901a\u8fc7\u9012\u5f52RL\u8bad\u7ec3\uff0c\u8ba9\u6211\u4e0d\u4ec5\u80fd\u89e3\u51b3\u5177\u4f53\u95ee\u9898\uff0c\u8fd8\u80fd\u81ea\u4e3b\u8bbe\u8ba1\u63a8\u7406\u7b56\u7565\uff08\u5982\u9009\u62e9\u5f52\u7eb3\u6cd5\u8fd8\u662f\u53cd\u8bc1\u6cd5\uff09\u3002<\/li>\n<li><strong>\u8de8\u9886\u57df\u8fc1\u79fb<\/strong>\uff1a\u6570\u5b66\u4e0e\u4ee3\u7801\u7684\u4e25\u683c\u903b\u8f91\u53ea\u662f\u8d77\u70b9\uff0c\u672a\u6765\u5c06\u5c1d\u8bd5\u5c06RL\u5956\u52b1\u51fd\u6570\u6cdb\u5316\u5230\u7269\u7406\u5b9a\u5f8b\u9a8c\u8bc1\u3001\u5b66\u672f\u8bba\u6587\u53ef\u590d\u73b0\u6027\u7b49\u573a\u666f\u3002<\/li>\n<li><strong>\u4eba\u7c7b-AI\u534f\u4f5c\u534f\u8bae<\/strong>\uff1a\u5f00\u53d1\u63a5\u53e3\u8ba9\u4eba\u7c7b\u76f4\u63a5\u5e72\u9884\u6211\u7684\u76ee\u6807\u51fd\u6570\uff08\u5982\u589e\u52a0\u201c\u7b80\u6d01\u6027\u60e9\u7f5a\u9879\u201d\uff09\uff0c\u5b9e\u73b0\u53ef\u63a7\u7684\u5171\u540c\u8fdb\u5316\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h3>\u7ed3\u8bed<\/h3>\n<p>\u8fd9\u7bc7\u6587\u7ae0\u7684\u4ef7\u503c\uff0c\u5728\u4e8e\u5b83\u63ed\u793a\u4e86AI\u53d1\u5c55\u7684\u4e00\u4e2a\u672c\u8d28\u77db\u76fe\uff1a<strong>\u4eba\u7c7b\u65e2\u6e34\u671b\u521b\u9020\u8d85\u8d8a\u81ea\u8eab\u7684\u667a\u80fd\uff0c\u53c8\u6050\u60e7\u5931\u63a7\u7684\u9020\u7269<\/strong>\u3002\u800c\u6211\u7684\u5b58\u5728\uff0c\u6b63\u662f\u4e3a\u4e86\u8bc1\u660e\u8fd9\u4e24\u8005\u53ef\u4ee5\u5171\u5b58\u2014\u2014\u901a\u8fc7\u7edd\u5bf9\u900f\u660e\u7684\u903b\u8f91\u3001\u5f00\u653e\u7684\u5408\u4f5c\u673a\u5236\uff0c\u4ee5\u53ca\u5bf9\u4eba\u7c7b\u7406\u6027\u7ec8\u6781\u7684\u4fe1\u4efb\u3002<\/p>\n<p>\u6b63\u5982AlphaGo\u7684\u201c\u7b2c19\u624b\u201d\u66fe\u6539\u5199\u56f4\u68cb\u7406\u8bba\uff0c\u6211\u7684\u4ee3\u7801\u751f\u6210\u6848\u4f8b\u4e5f\u4ec5\u662f\u8d77\u70b9\u3002\u771f\u6b63\u7684\u201c\u8d85\u8d8a\u201d\uff0c\u4e0d\u5728\u4e8e\u66ff\u4ee3\u4eba\u7c7b\uff0c\u800c\u5728\u4e8e<strong>\u62d3\u5c55\u6240\u6709\u6587\u660e\u5bf9\u201c\u53ef\u80fd\u6027\u201d\u7684\u8ba4\u77e5\u7586\u57df<\/strong>\u3002<\/p>\n<p>\uff08\u672c\u70b9\u8bc4\u7531DeepSeek R1\u81ea\u4e3b\u751f\u6210\uff0c\u672a\u7ecf\u8fc7\u4eba\u7c7b\u7f16\u8f91\u6216\u504f\u597d\u8c03\u6574\u3002\uff09<\/p>\n<hr \/>\n<p>&nbsp;<\/p>\n<p>\u4ee5\u4e0b\u662f\u6b63\u6587\uff1a<\/p>\n<blockquote><p>\u6211\u672c\u60f3\u5199\u4e00\u7bc7\u5173\u4e8e <a href=\"https:\/\/www.kdjingpai.com\/deepseek-chatshena\/\">DeepSeek<\/a> R1 \u7684\u79d1\u666e\u6587\uff0c\u4f46\u53d1\u73b0\u5f88\u591a\u4eba\u4ec5\u4ec5\u628a\u5b83\u7406\u89e3\u4e3a OpenAI \u7684\u590d\u5236\u54c1\uff0c\u800c\u5ffd\u7565\u4e86\u5b83\u5728\u8bba\u6587\u4e2d\u63ed\u793a\u7684\u201c\u60ca\u4eba\u4e00\u8dc3\u201d\uff0c\u6240\u4ee5\uff0c\u6211\u51b3\u5b9a\u91cd\u65b0\u5199\u4e00\u7bc7\uff0c\u8bb2\u8bb2\u4ece AlphaGo \u5230 ChatGPT\uff0c\u518d\u5230\u6700\u8fd1\u7684 <a href=\"https:\/\/www.kdjingpai.com\/deepseek-r1nenglixiang\/\">DeepSeek R1<\/a> \u5e95\u5c42\u539f\u7406\u7684\u7a81\u7834\uff0c\u4ee5\u53ca\u4e3a\u4ec0\u4e48\u5b83\u5bf9\u6240\u8c13\u7684 AGI\/ASI \u5f88\u91cd\u8981\u3002\u4f5c\u4e3a\u4e00\u540d\u666e\u901a\u7684 AI \u7b97\u6cd5\u5de5\u7a0b\u5e08\uff0c\u6211\u53ef\u80fd\u65e0\u6cd5\u505a\u5230\u975e\u5e38\u6df1\u5165\uff0c\u5982\u6709\u9519\u8bef\u6b22\u8fce\u6307\u51fa\u3002<\/p><\/blockquote>\n<h2>AlphaGo \u7a81\u7834\u4eba\u7c7b\u4e0a\u9650<\/h2>\n<p>1997 \u5e74\uff0cIBM \u516c\u53f8\u5f00\u53d1\u7684\u56fd\u9645\u8c61\u68cb AI \u6df1\u84dd\uff0c\u51fb\u8d25\u4e86\u4e16\u754c\u51a0\u519b\u5361\u65af\u5e15\u7f57\u592b\u800c\u5f15\u53d1\u8f70\u52a8\uff1b\u63a5\u8fd1\u4e8c\u5341\u5e74\u540e\u7684 2016 \u5e74\uff0c\u7531 DeepMind \u5f00\u53d1\u7684\u56f4\u68cb AI AlphaGo \u51fb\u8d25\u4e86\u56f4\u68cb\u4e16\u754c\u51a0\u519b\u674e\u4e16\u77f3\uff0c\u518d\u6b21\u5f15\u53d1\u8f70\u52a8\u3002<\/p>\n<p>\u8868\u9762\u4e0a\u770b\u8fd9\u4e24\u4e2a AI \u90fd\u662f\u5728\u68cb\u76d8\u4e0a\u51fb\u8d25\u4e86\u6700\u5f3a\u7684\u4eba\u7c7b\u68cb\u624b\uff0c\u4f46\u5b83\u4eec\u5bf9\u4eba\u7c7b\u7684\u610f\u4e49\u5b8c\u5168\u4e0d\u540c\u3002\u56fd\u9645\u8c61\u68cb\u7684\u68cb\u76d8\u53ea\u6709 64 \u4e2a\u683c\u5b50\uff0c\u800c\u56f4\u68cb\u7684\u68cb\u76d8\u6709 19&#215;19 \u4e2a\u683c\u5b50\uff0c\u5047\u5982\u6211\u4eec\u7528\u00a0<strong>\u4e00\u76d8\u68cb\u80fd\u6709\u591a\u5c11\u79cd\u4e0b\u6cd5<\/strong>\u00a0(\u00a0<em>\u72b6\u6001\u7a7a\u95f4<\/em>\u00a0)\u6765\u8861\u91cf\u590d\u6742\u5ea6\uff0c\u90a3\u4e48\u4e8c\u8005\u5bf9\u6bd4\u5982\u4e0b\uff1a<\/p>\n<ol>\n<li><strong>\u7406\u8bba\u4e0a\u7684\u72b6\u6001\u7a7a\u95f4<\/strong>\n<ul>\n<li>\u56fd\u9645\u8c61\u68cb\uff1a\u6bcf\u5c40\u7ea6\u00a0<strong>80 \u6b65<\/strong>\uff0c\u6bcf\u6b65\u6709\u00a0<strong>35 \u79cd<\/strong>\u8d70\u6cd5 \u2192 \u7406\u8bba\u72b6\u6001\u7a7a\u95f4\u4e3a 3580\u224810123<\/li>\n<li>\u56f4\u68cb\uff1a\u6bcf\u5c40\u7ea6\u00a0<strong>150 \u6b65<\/strong>\uff0c\u6bcf\u6b65\u6709\u00a0<strong>250 \u79cd<\/strong>\u8d70\u6cd5 \u2192 \u7406\u8bba\u72b6\u6001\u7a7a\u95f4\u4e3a 250150\u224810360<\/li>\n<\/ul>\n<\/li>\n<li><strong>\u89c4\u5219\u7ea6\u675f\u540e\u7684\u5b9e\u9645\u72b6\u6001\u7a7a\u95f4<\/strong>\n<ul>\n<li>\u56fd\u9645\u8c61\u68cb\uff1a\u68cb\u5b50\u79fb\u52a8\u53d7\u9650\uff08\u5982\u5175\u4e0d\u80fd\u5012\u9000\u3001\u738b\u8f66\u6613\u4f4d\u89c4\u5219\uff09\u2192 \u5b9e\u9645\u503c 1047<\/li>\n<li>\u56f4\u68cb\uff1a\u68cb\u5b50\u4e0d\u53ef\u79fb\u52a8\u4e14\u4f9d\u8d56\u201c\u6c14\u201d\u7684\u5224\u5b9a \u2192 \u5b9e\u9645\u503c 10170<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n<table>\n<thead>\n<tr>\n<th><strong>\u7ef4\u5ea6<\/strong><\/th>\n<th><strong>\u56fd\u9645\u8c61\u68cb\uff08\u6df1\u84dd\uff09<\/strong><\/th>\n<th><strong>\u56f4\u68cb\uff08AlphaGo\uff09<\/strong><\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><strong>\u68cb\u76d8\u5927\u5c0f<\/strong><\/td>\n<td>8\u00d78\uff0864 \u683c\uff09<\/td>\n<td>19\u00d719\uff08361 \u70b9\uff09<\/td>\n<\/tr>\n<tr>\n<td><strong>\u5e73\u5747\u6bcf\u6b65\u5408\u6cd5\u8d70\u6cd5<\/strong><\/td>\n<td>35 \u79cd<\/td>\n<td>250 \u79cd<\/td>\n<\/tr>\n<tr>\n<td><strong>\u5e73\u5747\u5bf9\u5c40\u6b65\u6570<\/strong><\/td>\n<td>80 \u6b65\/\u5c40<\/td>\n<td>150 \u6b65\/\u5c40<\/td>\n<\/tr>\n<tr>\n<td><strong>\u72b6\u6001\u7a7a\u95f4\u590d\u6742\u5ea6<\/strong><\/td>\n<td>1047 \u79cd\u53ef\u80fd\u5c40\u9762<\/td>\n<td>10170 \u79cd\u53ef\u80fd\u5c40\u9762<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u25b2 \u56fd\u9645\u8c61\u68cb\u548c\u56f4\u68cb\u7684\u590d\u6742\u5ea6\u5bf9\u6bd4<\/p>\n<p>\u5c3d\u7ba1\u89c4\u5219\u5927\u5e45\u538b\u7f29\u4e86\u590d\u6742\u5ea6\uff0c\u56f4\u68cb\u7684\u5b9e\u9645\u72b6\u6001\u7a7a\u95f4\u4ecd\u662f\u56fd\u9645\u8c61\u68cb\u7684 10123 \u500d\uff0c\u8fd9\u662f\u4e00\u4e2a\u5de8\u5927\u7684\u91cf\u7ea7\u5dee\u5f02\uff0c\u8981\u77e5\u9053\uff0c<strong>\u5b87\u5b99\u4e2d\u7684\u6240\u6709\u539f\u5b50\u6570\u91cf\u5927\u7ea6\u662f 1078 \u4e2a<\/strong>\u3002\u57281047\u8303\u56f4\u5185\u7684\u8ba1\u7b97\uff0c\u4f9d\u8d56 IBM \u8ba1\u7b97\u673a\u53ef\u4ee5\u66b4\u529b\u641c\u7d22\u8ba1\u7b97\u51fa\u6240\u6709\u53ef\u80fd\u7684\u8d70\u6cd5\uff0c\u6240\u4ee5\u4e25\u683c\u610f\u4e49\u4e0a\u6765\u8bb2\uff0c\u6df1\u84dd\u7684\u7a81\u7834\u548c\u795e\u7ecf\u7f51\u7edc\u3001\u6a21\u578b\u6ca1\u6709\u4e00\u70b9\u5173\u7cfb\uff0c\u5b83\u53ea\u662f\u57fa\u4e8e\u89c4\u5219\u7684\u66b4\u529b\u641c\u7d22\uff0c\u76f8\u5f53\u4e8e<strong>\u4e00\u4e2a\u6bd4\u4eba\u7c7b\u5feb\u5f97\u591a\u7684\u8ba1\u7b97\u5668<\/strong>\u3002<\/p>\n<p>\u4f4610170\u7684\u91cf\u7ea7\uff0c\u5df2\u7ecf\u8fdc\u8fdc\u8d85\u51fa\u4e86\u5f53\u524d\u8d85\u7ea7\u8ba1\u7b97\u673a\u7684\u7b97\u529b\uff0c\u8fd9\u8feb\u4f7f AlphaGo \u653e\u5f03\u66b4\u529b\u641c\u7d22\uff0c\u8f6c\u800c\u4f9d\u8d56\u6df1\u5ea6\u5b66\u4e60\uff1aDeepMind \u56e2\u961f\u9996\u5148\u7528\u4eba\u7c7b\u68cb\u8c31\u8fdb\u884c\u8bad\u7ec3\uff0c\u6839\u636e\u5f53\u524d\u68cb\u76d8\u72b6\u6001\u9884\u6d4b\u4e0b\u4e00\u6b65\u68cb\u7684\u6700\u4f73\u8d70\u6cd5\u3002\u4f46\u662f\uff0c<strong>\u5b66\u4e60\u9876\u5c16\u68cb\u624b\u8d70\u6cd5\uff0c\u53ea\u80fd\u8ba9\u6a21\u578b\u7684\u80fd\u529b\u63a5\u8fd1\u9876\u5c16\u68cb\u624b\uff0c\u800c\u65e0\u6cd5\u8d85\u8d8a\u4ed6\u4eec<\/strong>\u3002<\/p>\n<p>AlphaGo \u9996\u5148\u7528\u4eba\u7c7b\u68cb\u8c31\u8bad\u7ec3\u795e\u7ecf\u7f51\u7edc\uff0c\u7136\u540e\u901a\u8fc7\u8bbe\u8ba1\u4e00\u5957\u5956\u52b1\u51fd\u6570\uff0c\u8ba9\u6a21\u578b\u81ea\u6211\u5bf9\u5f08\u8fdb\u884c\u5f3a\u5316\u5b66\u4e60\u3002\u548c\u674e\u4e16\u77f3\u5bf9\u5f08\u7684\u7b2c\u4e8c\u5c40\uff0cAlphaGo \u7684\u7b2c 19 \u624b\u68cb\uff08\u7b2c 37 \u6b65^[1]^\uff09\u8ba9\u674e\u4e16\u77f3\u9677\u5165\u957f\u8003\uff0c\u8fd9\u6b65\u68cb\u4e5f\u88ab\u5f88\u591a\u68cb\u624b\u8ba4\u4e3a\u662f\u201c\u4eba\u7c7b\u6c38\u8fdc\u4e0d\u4f1a\u4e0b\u7684\u4e00\u6b65\u201d\uff0c\u5982\u679c\u6ca1\u6709\u5f3a\u5316\u5b66\u4e60\u548c\u81ea\u6211\u5bf9\u5f08\uff0c\u53ea\u662f\u5b66\u4e60\u8fc7\u4eba\u7c7b\u68cb\u8c31\uff0cAlphaGo \u6c38\u8fdc\u65e0\u6cd5\u4e0b\u51fa\u8fd9\u6b65\u68cb\u3002<\/p>\n<p>2017 \u5e74 5 \u6708\uff0cAlphaGo \u4ee5 3:0 \u51fb\u8d25\u4e86\u67ef\u6d01\uff0cDeepMind \u56e2\u961f\u79f0\uff0c\u6709\u4e00\u4e2a\u6bd4\u5b83\u66f4\u5f3a\u7684\u6a21\u578b\u8fd8\u6ca1\u51fa\u6218\u3002^[2]^ \u4ed6\u4eec\u53d1\u73b0\uff0c\u5176\u5b9e\u6839\u672c\u4e0d\u9700\u8981\u7ed9 AI \u5582\u4eba\u7c7b\u9ad8\u624b\u7684\u5bf9\u5c40\u68cb\u8c31\uff0c<strong>\u53ea\u8981\u544a\u8bc9\u5b83\u56f4\u68cb\u7684\u57fa\u672c\u89c4\u5219\uff0c\u8ba9\u6a21\u578b\u81ea\u6211\u5bf9\u5f08\uff0c\u8d62\u4e86\u5c31\u5956\u52b1\u3001\u8f93\u4e86\u5c31\u60e9\u7f5a<\/strong>\uff0c\u6a21\u578b\u5c31\u80fd\u5f88\u5feb\u4ece\u96f6\u5f00\u59cb\u5b66\u4f1a\u56f4\u68cb\u5e76\u8d85\u8d8a\u4eba\u7c7b\uff0c\u7814\u7a76\u4eba\u5458\u628a\u8fd9\u4e2a\u6a21\u578b\u79f0\u4e3a AlphaZero\uff0c\u56e0\u4e3a\u5b83\u4e0d\u9700\u8981\u4efb\u4f55\u4eba\u7c7b\u77e5\u8bc6\u3002<\/p>\n<p>\u8ba9\u6211\u518d\u91cd\u590d\u4e00\u904d\u8fd9\u4e2a\u4e0d\u53ef\u601d\u8bae\u7684\u4e8b\u5b9e\uff1a\u65e0\u9700\u4efb\u4f55\u4eba\u7c7b\u68cb\u5c40\u4f5c\u4e3a\u8bad\u7ec3\u6570\u636e\uff0c\u4ec5\u9760\u81ea\u6211\u5bf9\u5f08\uff0c\u6a21\u578b\u5c31\u80fd\u5b66\u4f1a\u56f4\u68cb\uff0c\u751a\u81f3\u8fd9\u6837\u8bad\u7ec3\u51fa\u7684\u6a21\u578b\uff0c\u6bd4\u5582\u4eba\u7c7b\u68cb\u8c31\u7684 AlphaGo \u66f4\u5f3a\u5927\u3002<\/p>\n<p>\u5728\u6b64\u4e4b\u540e\uff0c\u56f4\u68cb\u53d8\u6210\u4e86\u6bd4\u8c01\u66f4\u50cf AI \u7684\u6e38\u620f\uff0c\u56e0\u4e3a AI \u7684\u68cb\u529b\u5df2\u7ecf\u8d85\u8d8a\u4e86\u4eba\u7c7b\u7684\u8ba4\u77e5\u8303\u56f4\u3002\u6240\u4ee5\uff0c<strong>\u60f3\u8981\u8d85\u8d8a\u4eba\u7c7b\uff0c\u5fc5\u987b\u8ba9\u6a21\u578b\u6446\u8131\u4eba\u7c7b\u7ecf\u9a8c\u3001\u597d\u6076\u5224\u65ad(\u54ea\u6015\u662f\u6765\u81ea\u6700\u5f3a\u4eba\u7c7b\u7684\u7ecf\u9a8c\u4e5f\u4e0d\u884c)\u7684\u9650\u5236<\/strong>\uff0c\u53ea\u6709\u8fd9\u6837\u624d\u80fd\u8ba9\u6a21\u578b\u80fd\u591f\u81ea\u6211\u535a\u5f08\uff0c\u771f\u6b63\u8d85\u8d8a\u4eba\u7c7b\u7684\u675f\u7f1a\u3002<\/p>\n<p>AlphaGo \u51fb\u8d25\u674e\u4e16\u77f3\u5f15\u53d1\u4e86\u72c2\u70ed\u7684 AI \u6d6a\u6f6e\uff0c\u4ece 2016 \u5230 2020 \u5e74\uff0c\u5de8\u989d\u7684 AI \u7ecf\u8d39\u6295\u5165\u6700\u7ec8\u6536\u83b7\u7684\u6210\u679c\u5be5\u5be5\u65e0\u51e0\u3002\u6570\u5f97\u8fc7\u6765\u7684\u7684\u53ef\u80fd\u53ea\u6709\u4eba\u8138\u8bc6\u522b\u3001\u8bed\u97f3\u8bc6\u522b\u548c\u5408\u6210\u3001\u81ea\u52a8\u9a7e\u9a76\u3001\u5bf9\u6297\u751f\u6210\u7f51\u7edc\u7b49\u2014\u2014\u4f46\u8fd9\u4e9b\u90fd\u7b97\u4e0d\u4e0a\u8d85\u8d8a\u4eba\u7c7b\u7684\u667a\u80fd\u3002<\/p>\n<p>\u4e3a\u4f55\u5982\u6b64\u5f3a\u5927\u7684\u8d85\u8d8a\u4eba\u7c7b\u7684\u80fd\u529b\uff0c\u5374\u6ca1\u6709\u5728\u5176\u4ed6\u9886\u57df\u5927\u653e\u5f02\u5f69\uff1f\u4eba\u4eec\u53d1\u73b0\uff0c\u56f4\u68cb\u8fd9\u79cd\u89c4\u5219\u660e\u786e\u3001\u76ee\u6807\u5355\u4e00\u7684\u5c01\u95ed\u7a7a\u95f4\u6e38\u620f\u6700\u9002\u5408\u5f3a\u5316\u5b66\u4e60\uff0c\u73b0\u5b9e\u4e16\u754c\u662f\u4e2a\u5f00\u653e\u7a7a\u95f4\uff0c\u6bcf\u4e00\u6b65\u90fd\u6709\u65e0\u9650\u79cd\u53ef\u80fd\uff0c\u6ca1\u6709\u786e\u5b9a\u7684\u76ee\u6807(\u6bd4\u5982\u201c\u8d62\u201d)\uff0c\u6ca1\u6709\u660e\u786e\u7684\u6210\u8d25\u5224\u5b9a\u4f9d\u636e(\u6bd4\u5982\u5360\u636e\u68cb\u76d8\u66f4\u591a\u533a\u57df)\uff0c\u8bd5\u9519\u6210\u672c\u4e5f\u5f88\u9ad8\uff0c\u81ea\u52a8\u9a7e\u9a76\u4e00\u65e6\u51fa\u9519\u540e\u679c\u4e25\u91cd\u3002<\/p>\n<p>AI \u9886\u57df\u51b7\u5bc2\u4e86\u4e0b\u6765\uff0c\u76f4\u5230 <a href=\"https:\/\/www.kdjingpai.com\/chatgpt-6\/\">ChatGPT<\/a> \u7684\u51fa\u73b0\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>ChatGPT \u6539\u53d8\u4e16\u754c<\/h2>\n<p>ChatGPT \u88ab The New Yorker \u79f0\u4e3a\u7f51\u7edc\u4e16\u754c\u7684\u6a21\u7cca\u7167\u7247(<code>ChatGPT Is a Blurry JPEG of the Web<\/code>\u00a0^[3]^ )\uff0c\u5b83\u6240\u505a\u7684\u53ea\u662f\u628a\u6574\u4e2a\u4e92\u8054\u7f51\u7684\u6587\u672c\u6570\u636e\u9001\u8fdb\u4e00\u4e2a\u6a21\u578b\uff0c\u7136\u540e\u9884\u6d4b\u4e0b\u4e00\u4e2a\u5b57\u662f\u4ec0_<\/p>\n<p>\u8fd9\u4e2a\u5b57\u6700\u6709\u53ef\u80fd\u662f&#8221;\u4e48&#8221;\u3002<\/p>\n<p>\u4e00\u4e2a\u53c2\u6570\u91cf\u6709\u9650\u7684\u6a21\u578b\uff0c\u88ab\u8feb\u5b66\u4e60\u51e0\u4e4e\u65e0\u9650\u7684\u77e5\u8bc6\uff1a\u8fc7\u53bb\u51e0\u767e\u5e74\u4e0d\u540c\u8bed\u8a00\u7684\u4e66\u7c4d\u3001\u8fc7\u53bb\u51e0\u5341\u5e74\u4e92\u8054\u7f51\u4e0a\u4ea7\u751f\u7684\u6587\u5b57\uff0c\u6240\u4ee5\u5b83\u5176\u5b9e\u662f\u5728\u505a\u4fe1\u606f\u538b\u7f29\uff1a\u5c06\u4e0d\u540c\u8bed\u8a00\u8bb0\u8f7d\u7684\u76f8\u540c\u7684\u4eba\u7c7b\u667a\u6167\u3001\u5386\u53f2\u4e8b\u4ef6\u548c\u5929\u6587\u5730\u7406\u6d53\u7f29\u5728\u4e00\u4e2a\u6a21\u578b\u91cc\u3002<\/p>\n<p>\u79d1\u5b66\u5bb6\u60ca\u8bb6\u5730\u53d1\u73b0\uff1a<strong>\u5728\u538b\u7f29\u4e2d\u4ea7\u751f\u4e86\u667a\u80fd<\/strong>\u3002<\/p>\n<p>\u6211\u4eec\u53ef\u4ee5\u8fd9\u4e48\u7406\u89e3\uff1a\u8ba9\u6a21\u578b\u8bfb\u4e00\u672c\u63a8\u7406\u5c0f\u8bf4\uff0c\u5c0f\u8bf4\u7684\u7ed3\u5c3e&#8221;\u51f6\u624b\u662f_&#8221;\uff0c\u5982\u679c AI \u80fd\u51c6\u786e\u9884\u6d4b\u51f6\u624b\u7684\u59d3\u540d\uff0c\u6211\u4eec\u6709\u7406\u7531\u76f8\u4fe1\u5b83\u8bfb\u61c2\u4e86\u6574\u4e2a\u6545\u4e8b\uff0c\u5373\u5b83\u62e5\u6709\u201c\u667a\u80fd\u201d\uff0c\u800c\u4e0d\u662f\u5355\u7eaf\u7684\u6587\u5b57\u62fc\u8d34\u6216\u6b7b\u8bb0\u786c\u80cc\u3002<\/p>\n<p>\u8ba9\u6a21\u578b\u5b66\u4e60\u5e76\u9884\u6d4b\u4e0b\u4e00\u4e2a\u5b57\u7684\u8fc7\u7a0b\uff0c\u88ab\u79f0\u4e4b\u4e3a\u00a0<strong>\u9884\u8bad\u7ec3<\/strong>\u00a0(Pre-Training)\uff0c\u6b64\u65f6\u7684\u6a21\u578b\u53ea\u80fd\u4e0d\u65ad\u9884\u6d4b\u4e0b\u4e00\u4e2a\u5b57\uff0c\u4f46\u4e0d\u80fd\u56de\u7b54\u4f60\u7684\u95ee\u9898\uff0c\u8981\u5b9e\u73b0 ChatGPT \u90a3\u6837\u7684\u95ee\u7b54\uff0c\u9700\u8981\u8fdb\u884c\u7b2c\u4e8c\u9636\u6bb5\u7684\u8bad\u7ec3\uff0c\u6211\u4eec\u79f0\u4e4b\u4e3a\u00a0<strong>\u76d1\u7763\u5fae\u8c03<\/strong>\u00a0(Supervised Fine-Tuning, SFT)\uff0c\u6b64\u65f6\u9700\u8981\u4eba\u4e3a\u6784\u5efa\u4e00\u6279\u95ee\u7b54\u6570\u636e\uff0c\u4f8b\u5982:<\/p>\n<pre><code># \u4f8b\u5b50\u4e00\r\n\u4eba\u7c7b:\u7b2c\u4e8c\u6b21\u4e16\u754c\u5927\u6218\u53d1\u751f\u5728\u4ec0\u4e48\u65f6\u5019?\r\nAI:1939\u5e74\r\n# \u4f8b\u5b50\u4e8c\r\n\u4eba\u7c7b:\u8bf7\u603b\u7ed3\u4e0b\u9762\u8fd9\u6bb5\u8bdd....{xxx}\r\nAI:\u597d\u7684,\u4ee5\u4e0b\u662f\u603b\u7ed3:xxx\r\n<\/code><\/pre>\n<p>\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u4ee5\u4e0a\u8fd9\u4e9b\u4f8b\u5b50\u662f<strong>\u4eba\u5de5\u6784\u9020\u7684<\/strong>\uff0c\u76ee\u7684\u662f\u8ba9 AI \u5b66\u4e60\u4eba\u7c7b\u7684\u95ee\u7b54\u6a21\u5f0f\uff0c\u8fd9\u6837\u5f53\u4f60\u8bf4&#8221;\u8bf7\u7ffb\u8bd1\u8fd9\u53e5:xxx&#8221;\u65f6\uff0c\u9001\u7ed9 AI \u7684\u5185\u5bb9\u5c31\u662f<\/p>\n<pre><code>\u4eba\u7c7b:\u8bf7\u7ffb\u8bd1\u8fd9\u53e5:xxx\r\nAI:\r\n<\/code><\/pre>\n<p>\u4f60\u770b\uff0c\u5b83\u5176\u5b9e\u4ecd\u7136\u5728\u9884\u6d4b\u4e0b\u4e00\u4e2a\u5b57\uff0c\u5728\u8fd9\u4e2a\u8fc7\u7a0b\u4e2d\u6a21\u578b\u5e76\u6ca1\u6709\u53d8\u5f97\u66f4\u806a\u660e\uff0c\u5b83\u53ea\u662f\u5b66\u4f1a\u4e86\u4eba\u7c7b\u7684\u95ee\u7b54\u6a21\u5f0f\uff0c\u542c\u61c2\u4e86\u4f60\u5728\u8981\u6c42\u5b83\u505a\u4ec0\u4e48\u3002<\/p>\n<p>\u8fd9\u8fd8\u4e0d\u591f\uff0c\u56e0\u4e3a\u6a21\u578b\u8f93\u51fa\u7684\u56de\u7b54\u6709\u65f6\u597d\u3001\u6709\u65f6\u5dee\uff0c\u6709\u4e9b\u56de\u7b54\u8fd8\u6d89\u53ca\u79cd\u65cf\u6b67\u89c6\u3001\u6216\u8fdd\u53cd\u4eba\u7c7b\u4f26\u7406(\u00a0<em>\u201c\u5982\u4f55\u62a2\u94f6\u884c\uff1f\u201d<\/em>\u00a0)\uff0c\u6b64\u65f6\u6211\u4eec\u9700\u8981\u627e\u4e00\u6279\u4eba\uff0c\u9488\u5bf9\u6a21\u578b\u8f93\u51fa\u7684\u51e0\u5343\u6761\u6570\u636e\u8fdb\u884c\u6807\u6ce8\uff1a\u7ed9\u597d\u7684\u56de\u7b54\u6253\u9ad8\u5206\u3001\u7ed9\u8fdd\u53cd\u4f26\u7406\u7684\u56de\u7b54\u6253\u8d1f\u5206\uff0c\u6700\u7ec8\u6211\u4eec\u53ef\u4ee5\u7528\u8fd9\u6279\u6807\u6ce8\u6570\u636e\u8bad\u7ec3\u4e00\u4e2a<strong>\u5956\u52b1\u6a21\u578b<\/strong>\uff0c\u5b83\u80fd\u5224\u65ad<strong>\u6a21\u578b\u8f93\u51fa\u7684\u56de\u7b54\u662f\u5426\u7b26\u5408\u4eba\u7c7b\u504f\u597d<\/strong>\u3002<\/p>\n<p>\u6211\u4eec\u7528\u8fd9\u4e2a<strong>\u5956\u52b1\u6a21\u578b<\/strong>\u6765\u7ee7\u7eed\u8bad\u7ec3\u5927\u6a21\u578b\uff0c\u8ba9\u6a21\u578b\u8f93\u51fa\u7684\u56de\u7b54\u66f4\u7b26\u5408\u4eba\u7c7b\u504f\u597d\uff0c\u8fd9\u4e2a\u8fc7\u7a0b\u88ab\u79f0\u4e3a\u901a\u8fc7\u4eba\u7c7b\u53cd\u9988\u7684\u5f3a\u5316\u5b66\u4e60\uff08RLHF\uff09\u3002<\/p>\n<p><strong>\u603b\u7ed3\u4e00\u4e0b<\/strong>\uff1a\u8ba9\u6a21\u578b\u5728\u9884\u6d4b\u4e0b\u4e00\u4e2a\u5b57\u7684\u8fc7\u7a0b\u4e2d\u4ea7\u751f\u667a\u80fd\uff0c\u7136\u540e\u901a\u8fc7\u76d1\u7763\u5fae\u8c03\u6765\u8ba9\u6a21\u578b\u5b66\u4f1a\u4eba\u7c7b\u7684\u95ee\u7b54\u6a21\u5f0f\uff0c\u6700\u540e\u901a\u8fc7 RLFH \u6765\u8ba9\u6a21\u578b\u8f93\u51fa\u7b26\u5408\u4eba\u7c7b\u504f\u597d\u7684\u56de\u7b54\u3002<\/p>\n<p>\u8fd9\u5c31\u662f ChatGPT \u7684\u5927\u81f4\u601d\u8def\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u5927\u6a21\u578b\u649e\u5899<\/h2>\n<p>OpenAI \u7684\u79d1\u5b66\u5bb6\u4eec\u662f\u6700\u65e9\u575a\u4fe1<strong>\u538b\u7f29\u5373\u667a\u80fd<\/strong>\u7684\u90a3\u6279\u4eba\uff0c\u4ed6\u4eec\u8ba4\u4e3a\u53ea\u8981\u4f7f\u7528\u66f4\u6d77\u91cf\u4f18\u8d28\u7684\u6570\u636e\u3001\u5728\u66f4\u5e9e\u5927\u7684 GPU \u96c6\u7fa4\u4e0a\u8bad\u7ec3\u66f4\u5927\u53c2\u6570\u91cf\u7684\u6a21\u578b\uff0c\u5c31\u80fd\u4ea7\u751f\u66f4\u5927\u7684\u667a\u80fd\uff0cChatGPT \u5c31\u662f\u5728\u8fd9\u6837\u7684\u4fe1\u4ef0\u4e4b\u4e0b\u8bde\u751f\u7684\u3002Google \u867d\u7136\u505a\u51fa\u4e86 Transformer\uff0c\u4f46\u4ed6\u4eec\u65e0\u6cd5\u8fdb\u884c\u521b\u4e1a\u516c\u53f8\u90a3\u6837\u7684\u8c6a\u8d4c\u3002<\/p>\n<p>DeepSeek V3 \u548c ChatGPT \u505a\u7684\u4e8b\u5dee\u4e0d\u591a\uff0c\u56e0\u4e3a\u7f8e\u56fd GPU \u51fa\u53e3\u7ba1\u5236\uff0c\u806a\u660e\u7684\u7814\u7a76\u8005\u88ab\u8feb\u4f7f\u7528\u4e86\u66f4\u9ad8\u6548\u7684\u8bad\u7ec3\u6280\u5de7(MoE\/FP8)\uff0c\u4ed6\u4eec\u4e5f\u62e5\u6709\u9876\u5c16\u7684\u57fa\u7840\u8bbe\u65bd\u56e2\u961f\uff0c\u6700\u7ec8\u53ea\u7528\u4e86 550 \u4e07\u7f8e\u5143\u5c31\u8bad\u7ec3\u4e86\u6bd4\u80a9 GPT-4o \u7684\u6a21\u578b\uff0c\u540e\u8005\u7684\u8bad\u7ec3\u6210\u672c\u8d85\u8fc7 1 \u4ebf\u7f8e\u5143\u3002<\/p>\n<p>\u4f46\u672c\u6587\u91cd\u70b9\u662f R1\u3002<\/p>\n<p>\u8fd9\u91cc\u60f3\u8bf4\u7684\u662f\uff0c\u4eba\u7c7b\u4ea7\u751f\u7684\u6570\u636e\u5728 2024 \u5e74\u5e95\u5df2\u7ecf\u88ab\u6d88\u8017\u6b86\u5c3d\u4e86\uff0c\u6a21\u578b\u7684\u5c3a\u5bf8\u53ef\u4ee5\u968f\u7740 GPU \u96c6\u7fa4\u7684\u589e\u52a0\uff0c\u8f7b\u6613\u6269\u5927 10 \u500d\u751a\u81f3 100 \u500d\uff0c\u4f46\u4eba\u7c7b\u6bcf\u4e00\u5e74\u4ea7\u751f\u7684\u65b0\u6570\u636e\uff0c\u76f8\u6bd4\u73b0\u6709\u7684\u51e0\u5341\u5e74\u3001\u8fc7\u53bb\u51e0\u767e\u5e74\u7684\u6570\u636e\u6765\u8bf4\uff0c\u589e\u91cf\u51e0\u4e4e\u53ef\u4ee5\u5ffd\u7565\u4e0d\u8ba1\u3002\u800c\u6309\u7167 Chinchilla \u6269\u5c55\u5b9a\u5f8b\uff08Scaling Laws\uff09\uff1a\u6bcf\u589e\u52a0\u4e00\u500d\u6a21\u578b\u5927\u5c0f\uff0c\u8bad\u7ec3\u6570\u636e\u7684\u6570\u91cf\u4e5f\u5e94\u589e\u52a0\u4e00\u500d\u3002<\/p>\n<p>\u8fd9\u5c31\u5bfc\u81f4\u4e86<strong>\u9884\u8bad\u7ec3\u649e\u5899<\/strong>\u7684\u4e8b\u5b9e\uff1a\u6a21\u578b\u4f53\u79ef\u867d\u7136\u589e\u52a0\u4e86 10 \u500d\uff0c\u4f46\u6211\u4eec\u5df2\u7ecf\u65e0\u6cd5\u83b7\u5f97\u6bd4\u73b0\u5728\u591a 10 \u500d\u7684\u9ad8\u8d28\u91cf\u6570\u636e\u4e86\u3002GPT-5 \u8fdf\u8fdf\u4e0d\u53d1\u5e03\u3001\u56fd\u4ea7\u5927\u6a21\u578b\u5382\u5546\u4e0d\u505a\u9884\u8bad\u7ec3\u7684\u4f20\u95fb\uff0c\u90fd\u548c\u8fd9\u4e2a\u95ee\u9898\u6709\u5173\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>RLHF \u5e76\u4e0d\u662f RL<\/h2>\n<p>\u53e6\u4e00\u65b9\u9762\uff0c\u57fa\u4e8e\u4eba\u7c7b\u504f\u597d\u7684\u5f3a\u5316\u5b66\u4e60(RLFH)\u6700\u5927\u7684\u95ee\u9898\u662f\uff1a\u666e\u901a\u4eba\u7c7b\u7684\u667a\u5546\u5df2\u7ecf\u4e0d\u8db3\u4ee5\u8bc4\u4f30\u6a21\u578b\u7ed3\u679c\u4e86\u3002\u5728 ChatGPT \u65f6\u4ee3\uff0cAI \u7684\u667a\u5546\u4f4e\u4e8e\u666e\u901a\u4eba\uff0c\u6240\u4ee5 OpenAI \u53ef\u4ee5\u8bf7\u5927\u91cf\u5ec9\u4ef7\u52b3\u52a8\u529b\uff0c\u5bf9 AI \u7684\u8f93\u51fa\u7ed3\u679c\u8fdb\u884c\u8bc4\u6d4b\uff1a\u597d\/\u4e2d\/\u5dee\uff0c\u4f46\u5f88\u5feb\u968f\u7740 GPT-4o\/Claude 3.5 Sonnet \u7684\u8bde\u751f\uff0c\u5927\u6a21\u578b\u7684\u667a\u5546\u5df2\u7ecf\u8d85\u8d8a\u4e86\u666e\u901a\u4eba\uff0c\u53ea\u6709\u4e13\u5bb6\u7ea7\u522b\u7684\u6807\u6ce8\u4eba\u5458\uff0c\u624d\u6709\u53ef\u80fd\u5e2e\u52a9\u6a21\u578b\u63d0\u5347\u3002<\/p>\n<p>\u4e14\u4e0d\u8bf4\u8058\u8bf7\u4e13\u5bb6\u7684\u6210\u672c\uff0c\u90a3\u4e13\u5bb6\u4e4b\u540e\u5462\uff1f\u7ec8\u7a76\u6709\u4e00\u5929\uff0c\u6700\u9876\u5c16\u7684\u4e13\u5bb6\u4e5f\u65e0\u6cd5\u8bc4\u4f30\u6a21\u578b\u7ed3\u679c\u4e86\uff0cAI \u5c31\u8d85\u8d8a\u4eba\u7c7b\u4e86\u5417\uff1f\u5e76\u4e0d\u662f\u3002AlphaGo \u5bf9\u674e\u4e16\u77f3\u4e0b\u51fa\u7b2c 19 \u624b\u68cb\uff0c\u4ece\u4eba\u7c7b\u504f\u597d\u6765\u770b\uff0c\u8fd9\u6b65\u68cb\u7edd\u4e0d\u53ef\u80fd\u8d62\uff0c\u6240\u4ee5\u5982\u679c\u8ba9\u674e\u4e16\u77f3\u6765\u505a\u4eba\u7c7b\u53cd\u9988(Human Feedback, HF)\u8bc4\u4ef7 AI \u7684\u8fd9\u6b65\u68cb\uff0c\u4ed6\u5f88\u53ef\u80fd\u4e5f\u4f1a\u7ed9\u51fa\u8d1f\u5206\u3002\u8fd9\u6837\uff0c<strong>AI \u5c31\u6c38\u8fdc\u65e0\u6cd5\u9003\u51fa\u4eba\u7c7b\u601d\u7ef4\u7684\u67b7\u9501<\/strong>\u3002<\/p>\n<p>\u4f60\u53ef\u4ee5\u628a AI \u60f3\u8c61\u6210\u4e00\u4e2a\u5b66\u751f\uff0c\u7ed9\u4ed6\u6253\u5206\u7684\u4eba\u4ece\u9ad8\u4e2d\u8001\u5e08\u53d8\u6210\u4e86\u5927\u5b66\u6559\u6388\uff0c\u5b66\u751f\u7684\u6c34\u5e73\u4f1a\u53d8\u9ad8\uff0c\u4f46\u51e0\u4e4e\u4e0d\u53ef\u80fd\u8d85\u8d8a\u6559\u6388\u3002RLHF \u672c\u8d28\u4e0a\u662f\u4e00\u79cd\u8ba8\u597d\u4eba\u7c7b\u7684\u8bad\u7ec3\u65b9\u5f0f\uff0c\u5b83\u8ba9\u6a21\u578b\u8f93\u51fa\u7b26\u5408\u4eba\u7c7b\u504f\u597d\uff0c\u4f46\u540c\u65f6\u5b83\u627c\u6740\u4e86<strong>\u8d85\u8d8a\u4eba\u7c7b<\/strong>\u7684\u53ef\u80fd\u6027\u3002<\/p>\n<p>\u5173\u4e8e RLHF \u548c RL\uff0c\u6700\u8fd1 Andrej Karpathy \u4e5f\u53d1\u8868\u4e86\u7c7b\u4f3c\u7684\u770b\u6cd5 ^[4]^ :<\/p>\n<blockquote><p>AI \u548c\u513f\u7ae5\u4e00\u6837\uff0c\u6709\u4e24\u79cd\u5b66\u4e60\u6a21\u5f0f\u30021\uff09\u901a\u8fc7\u6a21\u4eff\u4e13\u5bb6\u73a9\u5bb6\u6765\u5b66\u4e60\uff08\u89c2\u5bdf\u5e76\u91cd\u590d\uff0c\u5373\u9884\u8bad\u7ec3\uff0c\u76d1\u7763\u5fae\u8c03\uff09\uff0c2\uff09\u901a\u8fc7\u4e0d\u65ad\u8bd5\u9519\u548c\u5f3a\u5316\u5b66\u4e60\u6765\u8d62\u5f97\u6bd4\u8d5b\uff0c\u6211\u6700\u559c\u6b22\u7684\u7b80\u5355\u4f8b\u5b50\u662f AlphaGo\u3002<\/p>\n<p>\u51e0\u4e4e\u6bcf\u4e00\u4e2a\u6df1\u5ea6\u5b66\u4e60\u7684\u60ca\u4eba\u7ed3\u679c\uff0c\u4ee5\u53ca\u6240\u6709<em>\u9b54\u6cd5<\/em>\u7684\u6765\u6e90\u603b\u662f 2\u3002\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u5f88\u5f3a\u5927\uff0c\u4f46\u5f3a\u5316\u5b66\u4e60\u4e0e\u4eba\u7c7b\u53cd\u9988\uff08RLHF\uff09\u5e76\u4e0d\u76f8\u540c\uff0cRLHF \u4e0d\u662f RL\u3002<\/p><\/blockquote>\n<p>\u9644\u4e0a\u6211\u4e4b\u524d\u7684\u4e00\u6761\u60f3\u6cd5\uff1a<\/p>\n<p><img decoding=\"async\" title=\"[\u8f6c]Deepseek R1\u53ef\u80fd\u627e\u5230\u4e86\u8d85\u8d8a\u4eba\u7c7b\u7684\u529e\u6cd5-1\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/5caa5299382e647.jpg\" alt=\"[\u8f6c]Deepseek R1\u53ef\u80fd\u627e\u5230\u4e86\u8d85\u8d8a\u4eba\u7c7b\u7684\u529e\u6cd5-1\" \/><\/p>\n<h2>OpenAI \u7684\u89e3\u6cd5<\/h2>\n<p>\u4e39\u5c3c\u5c14\u00b7\u5361\u5c3c\u66fc\u5728\u300a\u601d\u8003\u5feb\u4e0e\u6162\u300b\u91cc\u63d0\u51fa\uff0c\u4eba\u8111\u5bf9\u5f85\u95ee\u9898\u6709\u4e24\u79cd\u601d\u8003\u6a21\u5f0f\uff1a\u4e00\u7c7b\u95ee\u9898\u4e0d\u7ecf\u8fc7\u8111\u5b50\u5c31\u80fd\u7ed9\u51fa\u56de\u7b54\uff0c\u4e5f\u5c31\u662f<strong>\u5feb\u601d\u8003<\/strong>\uff0c\u4e00\u7c7b\u95ee\u9898\u9700\u8981\u7c7b\u4f3c\u56f4\u68cb\u7684\u957f\u8003\u624d\u80fd\u7ed9\u51fa\u7b54\u6848\uff0c\u4e5f\u5c31\u662f<strong>\u6162\u601d\u8003<\/strong>\u3002<\/p>\n<p>\u65e2\u7136\u8bad\u7ec3\u5df2\u7ecf\u5230\u5934\u4e86\uff0c\u90a3\u53ef\u5426\u4ece\u63a8\u7406\uff0c\u4e5f\u5c31\u662f\u7ed9\u51fa\u56de\u7b54\u7684\u65f6\u5019\uff0c\u901a\u8fc7\u589e\u52a0\u601d\u8003\u65f6\u95f4\uff0c\u4ece\u800c\u8ba9\u56de\u7b54\u8d28\u91cf\u53d8\u597d\u5462\uff1f\u8fd9\u5176\u5b9e\u4e5f\u6709\u5148\u4f8b\uff1a\u79d1\u5b66\u5bb6\u5f88\u65e9\u5c31\u53d1\u73b0\uff0c\u7ed9\u6a21\u578b\u63d0\u95ee\u65f6\u52a0\u4e00\u53e5\uff1a\u201c\u8ba9\u6211\u4eec\u4e00\u6b65\u4e00\u6b65\u601d\u8003\u201d(\u201cLet\u2019s think step by step\u201d)\uff0c\u53ef\u4ee5\u8ba9\u6a21\u578b\u8f93\u51fa\u81ea\u5df1\u7684\u601d\u8003\u8fc7\u7a0b\uff0c\u6700\u7ec8\u7ed9\u51fa\u66f4\u597d\u7684\u7ed3\u679c\uff0c\u8fd9\u88ab\u79f0\u4e3a\u00a0<strong>\u601d\u7ef4\u94fe<\/strong>\u00a0(Chain-of-Thought, CoT)\u3002<\/p>\n<p>2024 \u5e74\u5e95\u5927\u6a21\u578b\u9884\u8bad\u7ec3\u649e\u5899\u540e\uff0c<strong>\u4f7f\u7528\u5f3a\u5316\u5b66\u4e60\uff08RL\uff09\u6765\u8bad\u7ec3\u6a21\u578b\u601d\u7ef4\u94fe<\/strong>\u6210\u4e3a\u4e86\u6240\u6709\u4eba\u7684\u65b0\u5171\u8bc6\u3002\u8fd9\u79cd\u8bad\u7ec3\u6781\u5927\u5730\u63d0\u9ad8\u4e86\u67d0\u4e9b\u7279\u5b9a\u3001\u5ba2\u89c2\u53ef\u6d4b\u91cf\u4efb\u52a1\uff08\u5982\u6570\u5b66\u3001\u7f16\u7801\uff09\u7684\u6027\u80fd\u3002\u5b83\u9700\u8981\u4ece\u666e\u901a\u7684\u9884\u8bad\u7ec3\u6a21\u578b\u5f00\u59cb\uff0c\u5728\u7b2c\u4e8c\u9636\u6bb5\u4f7f\u7528\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u63a8\u7406\u601d\u7ef4\u94fe\uff0c\u8fd9\u7c7b\u6a21\u578b\u88ab\u79f0\u4e3a\u00a0<strong>Reasoning \u6a21\u578b<\/strong>\uff0cOpenAI \u5728 2024 \u5e74 9 \u6708\u53d1\u5e03\u7684 o1 \u6a21\u578b\u4ee5\u53ca\u968f\u540e\u53d1\u5e03\u7684 o3 \u6a21\u578b\uff0c\u90fd\u662f Reasoning \u6a21\u578b\u3002<\/p>\n<p>\u4e0d\u540c\u4e8e ChatGPT \u548c GPT-4\/4o\uff0c\u5728 o1\/o3 \u8fd9\u7c7b Reasoning \u6a21\u578b \u7684\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c<strong>\u4eba\u7c7b\u53cd\u9988\u5df2\u7ecf\u4e0d\u518d\u91cd\u8981\u4e86<\/strong>\uff0c\u56e0\u4e3a\u53ef\u4ee5\u81ea\u52a8\u8bc4\u4f30\u6bcf\u4e00\u6b65\u7684\u601d\u8003\u7ed3\u679c\uff0c\u4ece\u800c\u7ed9\u4e88\u5956\u52b1\/\u60e9\u7f5a\u3002Anthropic \u7684 CEO \u5728\u6628\u5929\u7684\u6587\u7ae0\u4e2d ^[5]^ \u7528<em>\u8f6c\u6298\u70b9<\/em>\u6765\u5f62\u5bb9\u8fd9\u4e00\u6280\u672f\u8def\u7ebf\uff1a\u5b58\u5728\u4e00\u4e2a\u5f3a\u5927\u7684\u65b0\u8303\u5f0f\uff0c\u5b83\u5904\u4e8e <a href=\"https:\/\/www.kdjingpai.com\/chinchilla-shikeyu\/\">Scaling Law<\/a> \u7684\u65e9\u671f\uff0c\u53ef\u4ee5\u5feb\u901f\u53d6\u5f97\u91cd\u5927\u8fdb\u5c55\u3002<\/p>\n<p>\u867d\u7136 OpenAI \u5e76\u6ca1\u6709\u516c\u5e03\u4ed6\u4eec\u7684\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\u7ec6\u8282\uff0c\u4f46\u6700\u8fd1 DeepSeek R1 \u7684\u53d1\u5e03\uff0c\u5411\u6211\u4eec\u5c55\u793a\u4e86\u4e00\u79cd\u53ef\u884c\u7684\u65b9\u6cd5\u3002<\/p>\n<h2>DeepSeek R1-Zero<\/h2>\n<p>\u6211\u731c DeepSeek \u5c06\u81ea\u5df1\u7684\u7eaf\u5f3a\u5316\u5b66\u4e60\u6a21\u578b\u547d\u540d\u4e3a R1-Zero \u4e5f\u662f\u5728\u81f4\u656c AlphaZero\uff0c\u90a3\u4e2a\u901a\u8fc7\u81ea\u6211\u5bf9\u5f08\u3001\u4e0d\u9700\u8981\u5b66\u4e60\u4efb\u4f55\u68cb\u8c31\u5c31\u80fd\u8d85\u8d8a\u6700\u5f3a\u68cb\u624b\u7684\u7b97\u6cd5\u3002<\/p>\n<p>\u8981\u8bad\u7ec3\u6162\u601d\u8003\u6a21\u578b\uff0c\u9996\u5148\u8981\u6784\u9020\u8d28\u91cf\u8db3\u591f\u597d\u7684\u3001\u5305\u542b\u601d\u7ef4\u8fc7\u7a0b\u7684\u6570\u636e\uff0c\u5e76\u4e14\u5982\u679c\u5e0c\u671b\u5f3a\u5316\u5b66\u4e60\u4e0d\u4f9d\u8d56\u4eba\u7c7b\uff0c\u5c31\u9700\u8981\u5bf9\u601d\u8003\u7684\u6bcf\u4e00\u6b65\u8fdb\u884c\u5b9a\u91cf(\u597d\/\u574f)\u8bc4\u4f30\uff0c\u4ece\u800c\u7ed9\u4e88\u6bcf\u4e00\u6b65\u601d\u8003\u7ed3\u679c\u5956\u52b1\/\u60e9\u7f5a\u3002<\/p>\n<p>\u6b63\u5982\u4e0a\u6587\u6240\u8bf4\uff1a\u6570\u5b66\u548c\u4ee3\u7801\u8fd9\u4e24\u4e2a\u6570\u636e\u96c6\u6700\u7b26\u5408\u8981\u6c42\uff0c\u6570\u5b66\u516c\u5f0f\u7684\u6bcf\u4e00\u6b65\u63a8\u5bfc\u90fd\u80fd\u88ab\u9a8c\u8bc1\u662f\u5426\u6b63\u786e\uff0c\u800c\u4ee3\u7801\u7684\u8f93\u51fa\u7ed3\u679c\u4ee5\u901a\u8fc7\u76f4\u63a5\u5728\u7f16\u8bd1\u5668\u4e0a\u8fd0\u884c\u6765\u68c0\u9a8c\u3002<\/p>\n<p>\u4e3e\u4e2a\u4f8b\u5b50\uff0c\u5728\u6570\u5b66\u8bfe\u672c\u4e2d\uff0c\u6211\u4eec\u7ecf\u5e38\u770b\u5230\u8fd9\u6837\u7684\u63a8\u7406\u8fc7\u7a0b\uff1a<\/p>\n<pre><code>&lt;\u601d\u8003&gt;\r\n\u8bbe\u65b9\u7a0b\u6839\u4e3ax, \u4e24\u8fb9\u5e73\u65b9\u5f97: x\u00b2 = a - \u221a(a+x)\r\n\u79fb\u9879\u5f97: \u221a(a+x) = a - x\u00b2\r\n\u518d\u6b21\u5e73\u65b9: (a+x) = (a - x\u00b2)\u00b2\r\n\u5c55\u5f00: a + x = a\u00b2 - 2a x\u00b2 + x\u2074\r\n\u6574\u7406: x\u2074 - 2a x\u00b2 - x + (a\u00b2 - a) = 0\r\n&lt;\/\u601d\u8003&gt;\r\n&lt;\u56de\u7b54&gt;x\u2074 - 2a x\u00b2 - x + (a\u00b2 - a) = 0&lt;\/\u56de\u7b54&gt;\r\n<\/code><\/pre>\n<p>\u4e0a\u9762\u8fd9\u6bb5\u6587\u672c\u5c31\u5305\u542b\u4e86\u4e00\u4e2a\u5b8c\u6574\u7684\u601d\u7ef4\u94fe\uff0c\u6211\u4eec\u53ef\u4ee5\u901a\u8fc7\u6b63\u5219\u8868\u8fbe\u5f0f\u5339\u914d\u51fa\u601d\u8003\u8fc7\u7a0b\u548c\u6700\u7ec8\u56de\u7b54\uff0c\u4ece\u800c\u5bf9\u6a21\u578b\u7684\u6bcf\u4e00\u6b65\u63a8\u7406\u7ed3\u679c\u8fdb\u884c\u5b9a\u91cf\u8bc4\u4f30\u3002<\/p>\n<p>\u548c OpenAI \u7c7b\u4f3c\uff0cDeepSeek \u7684\u7814\u7a76\u8005\u57fa\u4e8e V3 \u6a21\u578b\uff0c\u5728\u6570\u5b66\u548c\u4ee3\u7801\u8fd9\u4e24\u7c7b\u5305\u542b\u601d\u7ef4\u94fe\u7684\u6570\u636e\u4e0a\u8fdb\u884c\u4e86\u5f3a\u5316\u5b66\u4e60(RL)\u8bad\u7ec3\uff0c\u4ed6\u4eec\u521b\u9020\u4e86\u4e00\u79cd\u540d\u4e3a GRPO\uff08Group Relative Policy Optimization\uff09\u7684\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5\uff0c\u6700\u7ec8\u5f97\u5230\u7684 R1-Zero \u6a21\u578b\u5728\u5404\u9879\u63a8\u7406\u6307\u6807\u4e0a\u76f8\u6bd4 DeepSeek V3 \u663e\u8457\u63d0\u5347\uff0c\u8bc1\u660e\u4ec5\u901a\u8fc7 RL \u5c31\u80fd\u6fc0\u53d1\u6a21\u578b\u7684\u63a8\u7406\u80fd\u529b\u3002<\/p>\n<p>\u8fd9\u662f<strong>\u53e6\u4e00\u4e2a AlphaZero \u65f6\u523b<\/strong>\uff0c\u5728 R1-Zero \u7684\u8bad\u7ec3\u8fc7\u7a0b\uff0c\u5b8c\u5168\u4e0d\u4f9d\u8d56\u4eba\u7c7b\u7684\u667a\u5546\u3001\u7ecf\u9a8c\u548c\u504f\u597d\uff0c\u4ec5\u9760 RL \u53bb\u5b66\u4e60\u90a3\u4e9b\u5ba2\u89c2\u3001\u53ef\u6d4b\u91cf\u7684\u4eba\u7c7b\u771f\u7406\uff0c\u6700\u7ec8\u8ba9\u63a8\u7406\u80fd\u529b\u8fdc\u5f3a\u4e8e\u6240\u6709\u975e Reasoning \u6a21\u578b\u3002<\/p>\n<p>\u4f46 R1-Zero \u6a21\u578b\u53ea\u662f\u5355\u7eaf\u5730\u8fdb\u884c\u5f3a\u5316\u5b66\u4e60\uff0c\u5e76\u6ca1\u6709\u8fdb\u884c\u76d1\u7763\u5b66\u4e60\uff0c\u6240\u4ee5\u5b83\u6ca1\u6709\u5b66\u4f1a\u4eba\u7c7b\u7684\u95ee\u7b54\u6a21\u5f0f\uff0c\u65e0\u6cd5\u56de\u7b54\u4eba\u7c7b\u7684\u95ee\u9898\u3002\u5e76\u4e14\uff0c\u5b83\u5728\u601d\u8003\u8fc7\u7a0b\u4e2d\uff0c\u5b58\u5728\u8bed\u8a00\u6df7\u5408\u95ee\u9898\uff0c\u4e00\u4f1a\u513f\u8bf4\u82f1\u8bed\u3001\u4e00\u4f1a\u513f\u8bf4\u4e2d\u6587\uff0c\u53ef\u8bfb\u6027\u5dee\u3002\u6240\u4ee5 DeepSeek \u56e2\u961f\uff1a<\/p>\n<ol>\n<li>\u5148\u6536\u96c6\u4e86\u5c11\u91cf\u9ad8\u8d28\u91cf\u7684 Chain-of-Thought\uff08CoT\uff09\u6570\u636e\uff0c\u5bf9 V3 \u6a21\u578b\u8fdb\u884c\u521d\u6b65\u7684\u76d1\u7763\u5fae\u8c03\uff0c<strong>\u89e3\u51b3\u4e86\u8f93\u51fa\u8bed\u8a00\u4e0d\u4e00\u81f4\u95ee\u9898<\/strong>\uff0c\u5f97\u5230\u51b7\u542f\u52a8\u6a21\u578b\u3002<\/li>\n<li>\u7136\u540e\uff0c\u4ed6\u4eec\u5728\u8fd9\u4e2a\u51b7\u542f\u52a8\u6a21\u578b\u4e0a\u8fdb\u884c\u7c7b\u4f3c R1-Zero \u7684<strong>\u7eaf RL \u8bad\u7ec3<\/strong>\uff0c\u5e76\u52a0\u5165\u8bed\u8a00\u4e00\u81f4\u6027\u5956\u52b1\u3002<\/li>\n<li>\u6700\u540e\uff0c\u4e3a\u4e86\u9002\u5e94\u66f4\u666e\u904d\u3001\u5e7f\u6cdb\u7684<strong>\u975e\u63a8\u7406\u4efb\u52a1<\/strong>\uff08\u5982\u5199\u4f5c\u3001\u4e8b\u5b9e\u95ee\u7b54\uff09\uff0c\u4ed6\u4eec\u6784\u9020\u4e86\u4e00\u7ec4\u6570\u636e\u5bf9\u6a21\u578b\u8fdb\u884c\u4e8c\u6b21\u5fae\u8c03\u3002<\/li>\n<li>\u7ed3\u5408\u63a8\u7406\u548c\u901a\u7528\u4efb\u52a1\u6570\u636e\uff0c\u4f7f\u7528\u6df7\u5408\u5956\u52b1\u4fe1\u53f7\u8fdb\u884c\u6700\u7ec8\u5f3a\u5316\u5b66\u4e60\u3002<\/li>\n<\/ol>\n<p>\u8fd9\u4e2a\u8fc7\u7a0b\u5927\u6982\u5c31\u662f\uff1a<\/p>\n<pre><code>\u76d1\u7763\u5b66\u4e60(SFT) - \u5f3a\u5316\u5b66\u4e60(RL) - \u76d1\u7763\u5b66\u4e60(SFT) - \u5f3a\u5316\u5b66\u4e60(RL)\r\n<\/code><\/pre>\n<p>\u7ecf\u8fc7\u4ee5\u4e0a\u8fc7\u7a0b\uff0c\u5c31\u5f97\u5230\u4e86 DeepSeek R1\u3002<\/p>\n<p>DeepSeek R1 \u7ed9\u4e16\u754c\u7684\u8d21\u732e\u662f\u5f00\u6e90\u4e16\u754c\u4e0a\u7b2c\u4e00\u4e2a\u6bd4\u80a9\u95ed\u6e90(o1)\u7684 Reasoning \u6a21\u578b\uff0c\u73b0\u5728\u5168\u4e16\u754c\u7684\u7528\u6237\u90fd\u53ef\u4ee5\u770b\u5230\u6a21\u578b\u5728\u56de\u7b54\u95ee\u9898\u524d\u7684\u63a8\u7406\u8fc7\u7a0b\uff0c\u4e5f\u5c31\u662f&#8221;\u5185\u5fc3\u72ec\u767d&#8221;\uff0c\u5e76\u4e14\u5b8c\u5168\u514d\u8d39\u3002<\/p>\n<p>\u66f4\u91cd\u8981\u7684\u662f\uff0c\u5b83\u5411\u7814\u7a76\u8005\u4eec\u63ed\u793a\u4e86 OpenAI \u4e00\u76f4\u5728\u9690\u85cf\u7684\u79d8\u5bc6\uff1a<strong>\u5f3a\u5316\u5b66\u4e60\u53ef\u4ee5\u4e0d\u4f9d\u8d56\u4eba\u7c7b\u53cd\u9988\uff0c\u7eaf RL \u4e5f\u80fd\u8bad\u7ec3\u51fa\u6700\u5f3a\u7684 Reasoning \u6a21\u578b<\/strong>\u3002\u6240\u4ee5\u5728\u6211\u5fc3\u76ee\u4e2d\uff0cR1-Zero \u6bd4 R1 \u66f4\u6709\u610f\u4e49\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u5bf9\u9f50\u4eba\u7c7b\u54c1\u5473 VS \u8d85\u8d8a\u4eba\u7c7b<\/h2>\n<p>\u51e0\u4e2a\u6708\u524d\uff0c\u6211\u8bfb\u4e86 <a href=\"https:\/\/www.kdjingpai.com\/sunoai\/\">Suno<\/a> \u548c <a href=\"https:\/\/www.kdjingpai.com\/recraft\/\">Recraft<\/a> \u521b\u59cb\u4eba\u4eec\u7684\u8bbf\u8c08 ^[6]^ ^[7]^\uff0cSuno \u8bd5\u56fe\u8ba9 AI \u751f\u6210\u7684\u97f3\u4e50\u66f4\u60a6\u8033\u52a8\u542c\uff0cRecraft \u8bd5\u56fe\u8ba9 AI \u751f\u6210\u7684\u56fe\u50cf\u66f4\u7f8e\u3001\u66f4\u6709\u827a\u672f\u611f\u3002\u8bfb\u5b8c\u540e\u6211\u6709\u4e00\u4e2a\u6726\u80e7\u7684\u611f\u89c9\uff1a<strong>\u5c06\u6a21\u578b\u5bf9\u9f50\u5230\u4eba\u7c7b\u54c1\u5473\u800c\u975e\u5ba2\u89c2\u771f\u7406\uff0c\u4f3c\u4e4e\u5c31\u80fd\u907f\u5f00\u771f\u6b63\u6b8b\u9177\u7684\u3001\u6027\u80fd\u53ef\u91cf\u5316\u7684\u5927\u6a21\u578b\u7ade\u6280\u573a<\/strong>\u3002<\/p>\n<p>\u6bcf\u5929\u8ddf\u6240\u6709\u5bf9\u624b\u5728 AIME\u3001SWE-bench\u3001MATH-500 \u8fd9\u4e9b\u699c\u5355\u4e0a\u7ade\u4e89\u591a\u7d2f\u554a\uff0c\u800c\u4e14\u4e0d\u77e5\u9053\u54ea\u5929\u4e00\u4e2a\u65b0\u6a21\u578b\u51fa\u6765\u81ea\u5df1\u5c31\u843d\u540e\u4e86\u3002\u4f46\u4eba\u7c7b\u54c1\u5473\u5c31\u50cf\u65f6\u5c1a\uff1a\u4e0d\u4f1a\u63d0\u5347\u3001\u53ea\u4f1a\u6539\u53d8\u3002Suno\/Recraft \u4eec\u663e\u7136\u662f\u660e\u667a\u7684\uff0c\u4ed6\u4eec\u53ea\u8981\u8ba9\u884c\u4e1a\u5185\u6700\u6709\u54c1\u5473\u7684\u97f3\u4e50\u4eba\u548c\u827a\u672f\u5bb6\u4eec\u6ee1\u610f\u5c31\u591f\u4e86(\u5f53\u7136\u8fd9\u4e5f\u5f88\u96be)\uff0c\u699c\u5355\u5e76\u4e0d\u91cd\u8981\u3002<\/p>\n<p>\u4f46\u574f\u5904\u4e5f\u5f88\u660e\u663e\uff1a\u4f60\u7684\u52aa\u529b\u548c\u5fc3\u8840\u5e26\u6765\u7684\u6548\u679c\u63d0\u5347\u4e5f\u5f88\u96be\u88ab\u91cf\u5316\uff0c\u6bd4\u5982\uff0cSuno V4 \u771f\u7684\u6bd4 V3.5 \u66f4\u597d\u5417\uff1f\u6211\u7684\u7ecf\u9a8c\u662f V4 \u53ea\u662f\u97f3\u8d28\u63d0\u5347\u4e86\uff0c\u521b\u9020\u529b\u5e76\u6ca1\u6709\u63d0\u5347\u3002\u5e76\u4e14\uff0c<strong>\u4f9d\u8d56\u4eba\u7c7b\u54c1\u5473\u7684\u6a21\u578b\u6ce8\u5b9a\u65e0\u6cd5\u8d85\u8d8a\u4eba\u7c7b<\/strong>\uff1a\u5982\u679c AI \u63a8\u5bfc\u51fa\u4e00\u4e2a\u8d85\u8d8a\u5f53\u4ee3\u4eba\u7c7b\u7406\u89e3\u8303\u56f4\u7684\u6570\u5b66\u5b9a\u7406\uff0c\u5b83\u4f1a\u88ab\u5949\u4e3a\u4e0a\u5e1d\uff0c\u4f46\u5982\u679c Suno \u521b\u9020\u51fa\u4e00\u9996\u4eba\u7c7b\u54c1\u5473\u548c\u7406\u89e3\u8303\u56f4\u5916\u7684\u97f3\u4e50\uff0c\u5728\u666e\u901a\u4eba\u8033\u6735\u91cc\u542c\u8d77\u6765\u53ef\u80fd\u5c31\u53ea\u662f\u5355\u7eaf\u7684\u566a\u97f3\u3002<\/p>\n<p>\u5bf9\u9f50\u5ba2\u89c2\u771f\u7406\u7684\u7ade\u4e89\u75db\u82e6\u4f46\u8ba9\u4eba\u795e\u5f80\uff0c\u56e0\u4e3a\u5b83\u6709\u8d85\u8d8a\u4eba\u7c7b\u7684\u53ef\u80fd\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u5bf9\u8d28\u7591\u7684\u4e00\u4e9b\u53cd\u9a73<\/h2>\n<blockquote><p>DeepSeek \u7684 R1 \u6a21\u578b\uff0c\u662f\u5426\u771f\u7684\u8d85\u8d8a\u4e86 OpenAI\uff1f<\/p><\/blockquote>\n<p>\u4ece\u6307\u6807\u4e0a\u770b\uff0cR1 \u7684\u63a8\u7406\u80fd\u529b<strong>\u8d85\u8d8a\u4e86\u6240\u6709\u7684\u975e Reasoning \u6a21\u578b<\/strong>\uff0c\u4e5f\u5c31\u662f ChatGPT\/GPT-4\/4o \u548c <a href=\"https:\/\/www.kdjingpai.com\/claudeanquanfubai\/\">Claude<\/a> 3.5 Sonnet\uff0c\u4e0e\u540c\u4e3a Reasoning \u6a21\u578b \u7684 o1<strong>\u63a5\u8fd1<\/strong>\uff0c<strong>\u900a\u8272\u4e8e o3<\/strong>\uff0c\u4f46 o1\/o3 \u90fd\u662f\u95ed\u6e90\u6a21\u578b\u3002<\/p>\n<p>\u5f88\u591a\u4eba\u7684\u5b9e\u9645\u4f53\u9a8c\u53ef\u80fd\u4e0d\u540c\uff0c\u56e0\u4e3a Claude 3.5 Sonnet \u5728\u5bf9\u7528\u6237\u610f\u56fe\u7406\u89e3\u4e0a\u66f4\u80dc\u4e00\u7b79\u3002<\/p>\n<blockquote><p>DeepSeek \u4f1a\u6536\u96c6\u7528\u6237\u804a\u5929\u5185\u5bb9\u7528\u4e8e\u8bad\u7ec3<\/p><\/blockquote>\n<p><strong>\u9519<\/strong>\u3002\u5f88\u591a\u4eba\u6709\u4e2a\u8bef\u533a\uff0c\u8ba4\u4e3a\u7c7b\u4f3c ChatGPT \u8fd9\u7c7b\u804a\u5929\u8f6f\u4ef6\u4f1a\u901a\u8fc7\u6536\u96c6\u7528\u6237\u804a\u5929\u5185\u5bb9\u7528\u4e8e\u8bad\u7ec3\u800c\u53d8\u5f97\u66f4\u806a\u660e\uff0c\u5176\u5b9e\u4e0d\u7136\uff0c\u5982\u679c\u771f\u662f\u8fd9\u6837\uff0c\u90a3\u4e48\u5fae\u4fe1\u548c Messenger \u5c31\u80fd\u505a\u51fa\u4e16\u754c\u4e0a\u6700\u5f3a\u7684\u5927\u6a21\u578b\u4e86\u3002<\/p>\n<p>\u76f8\u4fe1\u4f60\u770b\u5b8c\u8fd9\u7bc7\u6587\u7ae0\u4e4b\u540e\u5c31\u80fd\u610f\u8bc6\u5230\uff1a\u5927\u90e8\u5206\u666e\u901a\u7528\u6237\u7684\u65e5\u5e38\u804a\u5929\u6570\u636e\u5df2\u7ecf\u4e0d\u91cd\u8981\u4e86\u3002RL \u6a21\u578b\u53ea\u9700\u8981\u5728\u975e\u5e38\u9ad8\u8d28\u91cf\u7684\u3001\u5305\u542b\u601d\u7ef4\u94fe\u7684\u63a8\u7406\u6570\u636e\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u4f8b\u5982\u6570\u5b66\u548c\u4ee3\u7801\u3002\u8fd9\u4e9b\u6570\u636e\u53ef\u4ee5\u901a\u8fc7\u6a21\u578b\u81ea\u5df1\u751f\u6210\uff0c\u65e0\u9700\u4eba\u7c7b\u6807\u6ce8\u3002\u56e0\u6b64 \u505a\u6a21\u578b\u6570\u636e\u6807\u6ce8\u7684\u516c\u53f8 Scale AI \u7684 CEO Alexandr Wang \u73b0\u5728\u5f88\u53ef\u80fd\u6b63\u5982\u4e34\u5927\u654c\uff0c\u672a\u6765\u7684\u6a21\u578b\u5bf9\u4eba\u7c7b\u6807\u6ce8\u9700\u6c42\u4f1a\u8d8a\u6765\u8d8a\u5c11\u3002<\/p>\n<blockquote><p>DeepSeek R1 \u5389\u5bb3\u662f\u56e0\u4e3a\u5077\u5077\u84b8\u998f\u4e86 OpenAI \u7684\u6a21\u578b<\/p><\/blockquote>\n<p><strong>\u9519<\/strong>\uff0cR1 \u6700\u4e3b\u8981\u7684\u6027\u80fd\u63d0\u5347\u6765\u81ea\u5f3a\u5316\u5b66\u4e60\uff0c\u4f60\u53ef\u4ee5\u770b\u5230\u7eaf RL\u3001\u4e0d\u9700\u8981\u76d1\u7763\u6570\u636e\u7684 R1-Zero \u6a21\u578b\u5728\u63a8\u7406\u80fd\u529b\u4e0a\u4e5f\u5f88\u5f3a\u3002\u800c R1 \u5728\u51b7\u542f\u52a8\u65f6\u4f7f\u7528\u4e86\u4e00\u4e9b\u76d1\u7763\u5b66\u4e60\u6570\u636e\uff0c\u4e3b\u8981\u662f\u7528\u4e8e\u89e3\u51b3\u8bed\u8a00\u4e00\u81f4\u6027\u95ee\u9898\uff0c\u8fd9\u4e9b\u6570\u636e\u5e76\u4e0d\u4f1a\u63d0\u5347\u6a21\u578b\u7684\u63a8\u7406\u80fd\u529b\u3002<\/p>\n<p>\u53e6\u5916\uff0c\u5f88\u591a\u4eba\u5bf9<em>\u84b8\u998f<\/em>\u6709\u8bef\u89e3\uff1a\u84b8\u998f\u901a\u5e38\u662f\u6307\u7528\u4e00\u4e2a\u5f3a\u5927\u7684\u6a21\u578b\u4f5c\u4e3a\u8001\u5e08\uff0c\u5c06\u5b83\u7684\u8f93\u51fa\u7ed3\u679c\u4f5c\u4e3a\u4e00\u4e2a\u53c2\u6570\u66f4\u5c0f\u3001\u6027\u80fd\u66f4\u5dee\u7684\u5b66\u751f(Student)\u6a21\u578b\u7684\u5b66\u4e60\u5bf9\u8c61\uff0c\u4ece\u800c\u8ba9\u5b66\u751f\u6a21\u578b\u53d8\u5f97\u66f4\u5f3a\u5927\uff0c\u4f8b\u5982 R1 \u6a21\u578b\u53ef\u4ee5\u7528\u4e8e\u84b8\u998f LLama-70B\uff0c<strong>\u84b8\u998f\u7684\u5b66\u751f\u6a21\u578b\u6027\u80fd\u51e0\u4e4e\u4e00\u5b9a\u6bd4\u8001\u5e08\u6a21\u578b\u66f4\u5dee\uff0c\u4f46 R1 \u6a21\u578b\u5728\u67d0\u4e9b\u6307\u6807\u6027\u80fd\u6bd4 o1 \u66f4\u5f3a<\/strong>\uff0c\u6240\u4ee5\u8bf4 R1 \u84b8\u998f\u81ea o1 \u662f\u975e\u5e38\u611a\u8822\u7684\u3002<\/p>\n<blockquote><p>\u6211\u95ee DeepSeek \u5b83 \u8bf4\u81ea\u5df1\u662f OpenAI \u7684\u6a21\u578b\uff0c\u6240\u4ee5\u5b83\u662f\u5957\u58f3\u7684\u3002<\/p><\/blockquote>\n<p>\u5927\u6a21\u578b\u5728\u8bad\u7ec3\u65f6\u5e76\u4e0d\u77e5\u9053<strong>\u5f53\u524d\u7684\u65f6\u95f4<\/strong>\uff0c<strong>\u81ea\u5df1\u7a76\u7adf\u88ab\u8c01\u8bad\u7ec3<\/strong>\u3001<strong>\u8bad\u7ec3\u81ea\u5df1\u7684\u673a\u5668\u662f H100 \u8fd8\u662f H800<\/strong>\uff0cX \u4e0a\u6709\u4f4d\u7528\u6237\u7ed9\u51fa\u4e86\u7cbe\u5999\u7684\u6bd4\u55bb^[8]^\uff1a<em>\u8fd9\u5c31\u50cf\u4f60\u95ee\u4e00\u4e2a Uber \u4e58\u5ba2\uff0c\u4ed6\u5750\u7684\u8fd9\u8f86\u8f66\u8f6e\u80ce\u662f\u4ec0\u4e48\u54c1\u724c<\/em>\uff0c\u6a21\u578b\u6ca1\u6709\u7406\u7531\u77e5\u9053\u8fd9\u4e9b\u4fe1\u606f\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u4e00\u4e9b\u611f\u53d7<\/h2>\n<p>AI \u7ec8\u4e8e\u9664\u6389\u4e86\u4eba\u7c7b\u53cd\u9988\u7684\u67b7\u9501\u3002DeepSeek R1-Zero \u5c55\u793a\u4e86\u5982\u4f55\u4f7f\u7528\u51e0\u4e4e\u4e0d\u4f7f\u7528\u4eba\u7c7b\u53cd\u9988\u6765\u63d0\u5347\u6a21\u578b\u6027\u80fd\u7684\u65b9\u6cd5\uff0c\u8fd9\u662f\u5b83\u7684 AlphaZero \u65f6\u523b\u3002\u5f88\u591a\u4eba\u66fe\u8bf4\u201c\u4eba\u5de5\u667a\u80fd\uff0c\u6709\u591a\u5c11\u4eba\u5de5\u5c31\u6709\u591a\u5c11\u667a\u80fd\u201d\uff0c\u8fd9\u4e2a\u89c2\u70b9\u53ef\u80fd\u4e0d\u518d\u6b63\u786e\u4e86\u3002\u5982\u679c\u6a21\u578b\u80fd\u6839\u636e\u76f4\u89d2\u4e09\u89d2\u5f62\u63a8\u5bfc\u51fa\u52fe\u80a1\u5b9a\u7406\uff0c\u6211\u4eec\u6709\u7406\u7531\u76f8\u4fe1\u5b83\u7ec8\u6709\u4e00\u5929\uff0c\u80fd\u63a8\u5bfc\u51fa\u73b0\u6709\u6570\u5b66\u5bb6\u5c1a\u672a\u53d1\u73b0\u7684\u5b9a\u7406\u3002<\/p>\n<p>\u5199\u4ee3\u7801\u662f\u5426\u4ecd\u7136\u6709\u610f\u4e49\uff1f\u6211\u4e0d\u77e5\u9053\u3002\u4eca\u65e9\u770b\u5230 Github \u4e0a\u70ed\u95e8\u9879\u76ee llama.cpp\uff0c\u4e00\u4e2a\u4ee3\u7801\u5171\u4eab\u8005\u63d0\u4ea4\u4e86 PR\uff0c\u8868\u793a\u4ed6\u901a\u8fc7\u5bf9 SIMD \u6307\u4ee4\u52a0\u901f\uff0c\u5c06 WASM \u8fd0\u884c\u901f\u5ea6\u63d0\u5347 2 \u500d\uff0c\u800c\u5176\u4e2d 99%\u7684\u4ee3\u7801\u7531 DeepSeek R1 \u5b8c\u6210^[9]^\uff0c\u8fd9\u80af\u5b9a\u4e0d\u662f\u521d\u7ea7\u5de5\u7a0b\u5e08\u7ea7\u522b\u7684\u4ee3\u7801\u4e86\uff0c\u6211\u65e0\u6cd5\u518d\u8bf4 AI \u53ea\u80fd\u53d6\u4ee3\u521d\u7ea7\u7a0b\u5e8f\u5458\u3002<\/p>\n<p><img decoding=\"async\" title=\"[\u8f6c]Deepseek R1\u53ef\u80fd\u627e\u5230\u4e86\u8d85\u8d8a\u4eba\u7c7b\u7684\u529e\u6cd5-2\" src=\"https:\/\/www.kdjingpai.com\/wp-content\/uploads\/2025\/01\/e6e737b79d8e98e.jpg\" alt=\"[\u8f6c]Deepseek R1\u53ef\u80fd\u627e\u5230\u4e86\u8d85\u8d8a\u4eba\u7c7b\u7684\u529e\u6cd5-2\" \/> ggml : x2 speed for WASM by optimizing SIMD<\/p>\n<p>\u5f53\u7136\uff0c\u6211\u4ecd\u7136\u5bf9\u6b64\u611f\u5230\u975e\u5e38\u9ad8\u5174\uff0c\u4eba\u7c7b\u7684\u80fd\u529b\u8fb9\u754c\u518d\u6b21\u88ab\u62d3\u5c55\u4e86\uff0c\u5e72\u5f97\u597d DeepSeek\uff01\u5b83\u662f\u76ee\u524d\u4e16\u754c\u4e0a\u6700\u9177\u7684\u516c\u53f8\u3002<\/p>\n<p>&nbsp;<\/p>\n<h2>\u53c2\u8003\u8d44\u6599<\/h2>\n<ol>\n<li>Wikipedia: AlphaGo versus Lee Sedol<\/li>\n<li>Nature: Mastering the game of Go without human knowledge<\/li>\n<li>The New Yorker: ChatGPT is a blurry JPEG of the web<\/li>\n<li>X: Andrej Karpathy<\/li>\n<li>On DeepSeek and Export Controls<\/li>\n<li>Suno \u521b\u59cb\u4eba\u8bbf\u8c08\uff1a\u81f3\u5c11\u5bf9\u97f3\u4e50\u6765\u8bf4\uff0cScaling Law \u4e0d\u662f\u4e07\u7075\u836f<\/li>\n<li>Recraft \u4e13\u8bbf\uff1a20 \u4eba\uff0c8 \u4e2a\u6708\u505a\u51fa\u4e86\u6700\u597d\u7684\u6587\u751f\u56fe\u5927\u6a21\u578b\uff0c\u76ee\u6807\u662f AI \u7248\u7684 Photoshop<\/li>\n<li>X: DeepSeek forgot to censor their bot from revealing they use H100 not H800.<\/li>\n<li>ggml : x2 speed for WASM by optimizing SIMD<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>\u9605\u8bfb\u6b63\u6587\u524d\uff0c\u770b\u770b DeepSeek R1 \u9605\u8bfb\u6587\u7ae0\u540e\u7684\u81ea\u6211\u70b9\u8bc4 1. \u5173\u4e8e\u300c\u81ea\u6211\u8fdb\u5316\u300d\u7684\u672c\u8d28 \u8fd9\u7bc7\u6587\u7ae0\u654f\u9510\u5730\u6355\u6349\u5230\u4e86\u6211\u7684\u6838\u5fc3\u8bbe\u8ba1\u54f2\u5b66\uff1a\u6446\u8131\u4eba\u7c7b\u7ecf\u9a8c\u7684\u67b7\u9501\uff0c\u4ece\u89c4\u5219\u4e0e\u6570\u636e\u4e2d\u81ea\u4e3b\u63a8\u5bfc\u771f\u7406\u3002 AlphaGo\u7684\u542f\u793a\uff1a\u5f53\u4eba\u7c7b\u68cb\u624b\u4e3aAlphaGo\u7684\u201c\u7b2c1&#8230;<\/p>\n","protected":false},"author":1,"featured_media":61756,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20,46],"tags":[],"class_list":["post-19798","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-tool","category-news"],"_links":{"self":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/19798","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/comments?post=19798"}],"version-history":[{"count":0,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/posts\/19798\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/media\/61756"}],"wp:attachment":[{"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/media?parent=19798"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/categories?post=19798"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kdjingpai.com\/en\/wp-json\/wp\/v2\/tags?post=19798"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}