{"id":53774,"date":"2025-02-16T11:29:01","date_gmt":"2025-02-16T03:29:01","guid":{"rendered":"https:\/\/fwq.ai\/blog\/53774\/"},"modified":"2025-02-16T11:29:01","modified_gmt":"2025-02-16T03:29:01","slug":"deepseek-r1-zero-r1%e5%88%86%e6%9e%90","status":"publish","type":"post","link":"https:\/\/fwq.ai\/blog\/53774\/","title":{"rendered":"DeepSeek R1-Zero &#038; R1\u5206\u6790"},"content":{"rendered":"<p>\u6211\u4eec\u8fd8\u6ca1\u6709 AGI\uff0c\u521b\u65b0\u4ecd\u7136\u53d7\u5230\u9650\u5236\u2014\u2014\u6269\u5927\u7eaf LLM \u9884\u8bad\u7ec3\u5e76\u4e0d\u662f\u51fa\u8def\uff0c\u5c3d\u7ba1\u8fd9\u662f\u53bb\u5e74\u590f\u5929\u4eba\u5de5\u667a\u80fd\u884c\u4e1a\u7684\u4e3b\u6d41\u53d9\u4e8b\u548c\u4e3b\u6d41\u516c\u4f17\u89c2\u70b9\u3002<\/p>\n<p>\u53d9\u4e8b\u4e4b\u6240\u4ee5\u91cd\u8981\uff0c\u662f\u56e0\u4e3a\u5b83\u4eec\u6700\u7ec8\u4f1a\u63a8\u52a8\u7ecf\u6d4e\u6d3b\u52a8\uff0c\u5982\u6295\u8d44\u3001\u7814\u7a76\u91cd\u70b9\u3001\u8d44\u91d1\u3001\u5730\u7f18\u653f\u6cbb\u3001\u8d38\u6613\u7b49\u3002\u4f8b\u5982\uff0c\u5728 2023-24 \u5e74\uff0c\u65b0 LLM \u521d\u521b\u516c\u53f8\u7684\u6295\u8d44\u7ea6\u4e3a 200 \u4ebf\u7f8e\u5143\uff0c\u800c\u65b0 AGI \u521d\u521b\u516c\u53f8\u7684\u6295\u8d44\u4ec5\u4e3a\u7ea6 2 \u4ebf\u7f8e\u5143\u3002<\/p>\n<p>\u6211\u4eec\u4e8e\u53bb\u5e74 6 \u6708\u63a8\u51fa\u4e86 ARC Prize 2024\uff0c\u4ee5\u63d0\u9ad8\u4eba\u4eec\u5bf9 LLM \u6269\u5c55\u9650\u5236\u7684\u8ba4\u8bc6\uff0c\u5e76\u63a8\u52a8\u4e00\u4e2a\u6709\u7528\u7684\u57fa\u51c6 ARC-AGI-1 \u671d\u7740\u65b0\u7684\u65b9\u5411\u53d1\u5c55\uff0c\u5373\u8981\u6c42 AI \u7cfb\u7edf\u9002\u5e94\u65b0\u7684\u3001\u770b\u4e0d\u89c1\u7684\u95ee\u9898\uff0c\u800c\u4e0d\u662f\u4e25\u683c\u4f9d\u8d56\u8bb0\u5fc6\u3002<\/p>\n<h2>1\u3001DeepSeek R1\u8bad\u7ec3\u67b6\u6784<\/h2>\n<p>  R1 \u8bad\u7ec3\u67b6\u6784\u3002 <\/p>\n<p>\u4e0a\u5468\uff0cDeepSeek \u53d1\u5e03\u4e86\u4ed6\u4eec\u7684\u65b0 R1-Zero \u548c R1\u201c\u63a8\u7406\u673a\u201d\u7cfb\u7edf\uff0c\u53ef\u4e0e OpenAI \u7684 ARC-AGI-1 \u4e0a\u7684 o1 \u7cfb\u7edf\u76f8\u5ab2\u7f8e\u3002R1-Zero\u3001R1 \u548c o1\uff08\u4f4e\u8ba1\u7b97\uff09\u7684\u5f97\u5206\u90fd\u5728 15-20% \u5de6\u53f3\u2014\u2014\u76f8\u6bd4\u4e4b\u4e0b\uff0cGPT-4o \u7684\u5f97\u5206\u4e3a 5%\uff0c\u8fd9\u662f\u591a\u5e74\u6765\u7eaf LLM \u6269\u5c55\u7684\u5dc5\u5cf0\u3002\u6839\u636e\u672c\u5468\u7f8e\u56fd\u5e02\u573a\u7684\u53cd\u5e94\uff0c\u516c\u4f17\u4e5f\u5f00\u59cb\u4e86\u89e3\u7eaf LLM \u6269\u5c55\u7684\u5c40\u9650\u6027\u3002\u7136\u800c\uff0c\u516c\u4f17\u5bf9\u5373\u5c06\u5230\u6765\u7684\u63a8\u7406\u9700\u6c42\u4ecd\u7136\u666e\u904d\u4e0d\u4e86\u89e3\u3002<\/p>\n<p>2024 \u5e74 12 \u6708\uff0cOpenAI \u5ba3\u5e03\u4e86\u6211\u4eec\u9a8c\u8bc1\u7684\u7a81\u7834\u6027\u65b0 o3 \u7cfb\u7edf\u3002\u5b83\u5728\u4f4e\u8ba1\u7b97\u6a21\u5f0f\u4e0b\u5f97\u5206\u4e3a 76%\uff0c\u5728\u9ad8\u8ba1\u7b97\u6a21\u5f0f\u4e0b\u5f97\u5206\u4e3a 88%\u3002o3 \u7cfb\u7edf\u5c55\u793a\u4e86\u8ba1\u7b97\u673a\u9002\u5e94\u65b0\u5947\u95ee\u9898\u7684\u9996\u4e2a\u5b9e\u7528\u3001\u901a\u7528\u5b9e\u73b0\u3002<\/p>\n<p>\u5c3d\u7ba1 o3 \u51fb\u8d25 ARC-AGI-1 \u662f\u91cd\u5927\u79d1\u6280\u65b0\u95fb\uff0c\u4f46\u4e3b\u6d41\u5a92\u4f53\u5374\u57fa\u672c\u6ca1\u6709\u6ce8\u610f\u5230\u548c\u62a5\u9053\u3002<\/p>\n<p>\u8fd9\u5bf9\u4e8e\u4eba\u5de5\u667a\u80fd\u9886\u57df\u548c\u8ba1\u7b97\u673a\u79d1\u5b66\u6765\u8bf4\u662f\u4e00\u4e2a\u6781\u5176\u91cd\u8981\u7684\u65f6\u523b\uff0c\u8fd9\u4e9b\u7cfb\u7edf\u9700\u8981\u7814\u7a76\u3002\u4f46\u7531\u4e8e o1\/o3 \u7684\u5c01\u95ed\u6027\uff0c\u6211\u4eec\u4e0d\u5f97\u4e0d\u4f9d\u8d56\u731c\u6d4b\u3002\u591a\u4e8f\u4e86 ARC-AGI-1 \u4ee5\u53ca\u73b0\u5728\uff08\u51e0\u4e4e\uff09\u5f00\u6e90\u7684 R1-Zero \u548c R1\uff0c\u6211\u4eec\u53ef\u4ee5\u52a0\u6df1\u7406\u89e3\u3002\u7279\u522b\u662f\uff0cR1-Zero \u6bd4 R1 \u91cd\u8981\u5f97\u591a\u3002<\/p>\n<p>\u6211\u4eec\u8bf4\u201c\u51e0\u4e4e\u201d\uff0c\u662f\u56e0\u4e3a DeepSeek \u8fd8\u6ca1\u6709\u53d1\u5e03\u4e00\u79cd\u53ef\u91cd\u590d\u7684\u65b9\u6cd5\u6765\u4ece\u5934\u5f00\u59cb\u751f\u6210\u6a21\u578b\u6743\u91cd\u3002<\/p>\n<h2>2\u3001R1-Zero \u6d88\u9664\u4e86\u4eba\u4e3a\u74f6\u9888<\/h2>\n<p>\u5728\u6211\u4eec\u7684 o1 \u548c o3 \u5206\u6790\u4e2d\uff0c\u6211\u4eec\u63a8\u6d4b\u4e86\u8fd9\u4e9b\u63a8\u7406\u7cfb\u7edf\u662f\u5982\u4f55\u5de5\u4f5c\u7684\u3002\u5173\u952e\u601d\u60f3\uff1a<\/p>\n<ul>\n<li>\u4e3a\u95ee\u9898\u57df\u751f\u6210\u601d\u7ef4\u94fe (CoT)\u3002<\/li>\n<li>\u4f7f\u7528\u4eba\u7c7b\u4e13\u5bb6\uff08\u201c\u76d1\u7763\u5fae\u8c03\u201d\u6216 SFT\uff09\u548c\u81ea\u52a8\u5316\u673a\u5668\uff08\u201c\u5f3a\u5316\u5b66\u4e60\u201d\u6216 RL\uff09\u7684\u7ec4\u5408\u6765\u6807\u8bb0\u4e2d\u95f4 CoT \u6b65\u9aa4\u3002<\/li>\n<li>\u4f7f\u7528 (2) \u8bad\u7ec3\u57fa\u7840\u6a21\u578b\u3002<\/li>\n<li>\u5728\u6d4b\u8bd5\u65f6\uff0c\u4ece\u6d41\u7a0b\u6a21\u578b\u8fdb\u884c\u8fed\u4ee3\u63a8\u7406\u3002<\/li>\n<\/ul>\n<p>\u4e0b\u9762\u56de\u987e\u4e86\u7528\u4e8e\u8fed\u4ee3\u91c7\u6837\u7684\u6280\u672f\u4ee5\u53ca ARC-AGI-1 \u5206\u6570\uff1a<\/p>\n<table>\n<thead>\n<tr>\n<th>\u7cfb\u7edf<\/th>\n<th>ARC-AGI-1<\/th>\n<th>\u65b9\u6cd5<\/th>\n<th>\u5e73\u5747\u4ee4\u724c<\/th>\n<th>\u5e73\u5747\u6210\u672c<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>r1-zero<\/td>\n<td>14%<\/td>\n<td>\u65e0 SFT \/ \u65e0\u641c\u7d22<\/td>\n<td>11K<\/td>\n<td>$.11<\/td>\n<\/tr>\n<tr>\n<td>r1<\/td>\n<td>15.8%<\/td>\n<td>SFT \/ \u65e0\u641c\u7d22<\/td>\n<td>6K<\/td>\n<td>$.06<\/td>\n<\/tr>\n<tr>\n<td>o1 (\u4f4e)<\/td>\n<td>20.5%<\/td>\n<td>SFT \/ \u65e0\u641c\u7d22<\/td>\n<td>7K<\/td>\n<td>$.43<\/td>\n<\/tr>\n<tr>\n<td>o1 (\u4e2d)<\/td>\n<td>31%<\/td>\n<td>SFT \/ \u65e0\u641c\u7d22<\/td>\n<td>13K<\/td>\n<td>$.79<\/td>\n<\/tr>\n<tr>\n<td>o1 (\u9ad8)<\/td>\n<td>35%<\/td>\n<td>SFT \/ \u65e0\u641c\u7d22<\/td>\n<td>22K<\/td>\n<td>$1.31<\/td>\n<\/tr>\n<tr>\n<td>o3 (\u4f4e)<\/td>\n<td>75.7%<\/td>\n<td>SFT \/ \u641c\u7d22 + \u91c7\u6837<\/td>\n<td>335K<\/td>\n<td>$20<\/td>\n<\/tr>\n<tr>\n<td>o3 (\u9ad8)<\/td>\n<td>87.5%<\/td>\n<td>SFT \/ \u641c\u7d22 + \u91c7\u6837<\/td>\n<td>57M<\/td>\n<td>$3.4K<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<blockquote>\n<p>\u6ce8\u610f\uff1a\u663e\u793a ARC-AGI-1 \u534a\u79c1\u6709\u5206\u6570\u3002<\/p>\n<\/blockquote>\n<p>\u501f\u52a9 DeepSeek \u6700\u65b0\u53d1\u8868\u7684\u7814\u7a76\uff0c\u6211\u4eec\u53ef\u4ee5\u66f4\u597d\u5730\u8fdb\u884c\u63a8\u6d4b\u3002\u5173\u952e\u89c1\u89e3\u662f\uff0cLLM \u63a8\u7406\u7cfb\u7edf\u66f4\u9ad8\u7a0b\u5ea6\u7684\u65b0\u9896\u6027\u9002\u5e94\u6027\uff08\u548c\u53ef\u9760\u6027\uff09\u662f\u5728\u4e09\u4e2a\u7ef4\u5ea6\u4e0a\u5b9e\u73b0\u7684\uff1a<\/p>\n<ul>\n<li>\u5c06\u4eba\u5de5\u6807\u7b7e\uff08\u53c8\u79f0 SFT\uff09\u6dfb\u52a0\u5230 CoT \u8fc7\u7a0b\u6a21\u578b\u8bad\u7ec3\u4e2d<\/li>\n<li>CoT \u641c\u7d22\u800c\u4e0d\u662f\u7ebf\u6027\u63a8\u7406\uff08\u5e76\u884c\u6bcf\u6b65 CoT \u63a8\u7406\uff09<\/li>\n<li>\u6574\u4e2a CoT \u91c7\u6837\uff08\u5e76\u884c\u8f68\u8ff9\u63a8\u7406\uff09<\/li>\n<\/ul>\n<p>\u9879\u76ee (1) \u53d7\u5230\u4eba\u5de5\u6570\u636e\u751f\u6210\u7684\u9650\u5236\uff0c\u5e76\u9650\u5236\u4e86\u8fd9\u4e9b\u63a8\u7406\u7cfb\u7edf\u6700\u53d7\u76ca\u7684\u9886\u57df\u3002\u4f8b\u5982\uff0cMMLU \u4e13\u4e1a\u6cd5\u5f8b\u7c7b\u522b\u4ee4\u4eba\u60ca\u8bb6\u5730\u6bd4 o1 \u4e0a\u7684\u6570\u5b66\u548c\u903b\u8f91\u4f4e\u5f97\u591a\u3002<\/p>\n<p>\u9879\u76ee (2) \u548c (3) \u53d7\u5230\u6548\u7387\u7684\u9650\u5236\u3002o1 \u548c o3 \u90fd\u663e\u793a\u51fa\u5bf9\u6570\u7ea7\u7684 ARC-AGI-1 \u57fa\u51c6\u6d4b\u8bd5\u51c6\u786e\u5ea6\u7684\u63d0\u9ad8\uff0c\u56e0\u4e3a\u5b83\u4eec\u5728\u6d4b\u8bd5\u65f6\u82b1\u8d39\u4e86\u66f4\u591a\u7684\u63a8\u7406\u8ba1\u7b97\uff0c\u800c\u82b1\u8d39\u8fd9\u4e9b\u8ba1\u7b97\u7684\u4e0d\u540c\u65b9\u5f0f\u4f1a\u8c03\u6574\u66f2\u7ebf\u7684 x \u8f74\u3002<\/p>\n<p>\u6211\u8ba4\u4e3a\uff0cDeepSeek \u6240\u505a\u7684\u6700\u6709\u8da3\u7684\u4e8b\u60c5\u662f\u5355\u72ec\u53d1\u5e03 R1-Zero\u3002R1-Zero \u662f\u4e00\u4e2a\u4e0d\u4f7f\u7528 SFT\uff08(1) \u9879\uff09\u7684\u6a21\u578b\u3002\u76f8\u53cd\uff0c\u5b83\u5b8c\u5168\u4f9d\u8d56\u4e8e\u5f3a\u5316\u5b66\u4e60\u3002<\/p>\n<p>R1-Zero \u548c R1 \u5728 ARC-AGI-1 \u4e0a\u8868\u73b0\u51fa\u5f88\u5f3a\u7684\u5f97\u5206\u4e00\u81f4\u6027\uff0c\u5206\u522b\u5f97\u5206\u4e3a 14% \u548c 15%\u3002DeepSeeks \u81ea\u5df1\u62a5\u544a\u7684\u57fa\u51c6\u5206\u6570\u4e5f\u663e\u793a\u51fa R1-Zero \u548c R1 \u4e4b\u95f4\u7684\u9ad8\u5ea6\u4e00\u81f4\u6027\uff0c\u4f8b\u5982\u5728 MATH AIME 2024 \u4e0a\u7684\u5f97\u5206\u5206\u522b\u4e3a 71% \u548c 76%\uff08\u9ad8\u4e8e\u57fa\u7840 DeepSeek V3 \u4e0a\u7684\u7ea6 40%\uff09\u3002<\/p>\n<p>\u5728\u8bba\u6587\u4e2d\uff0cR1-Zero \u7684\u4f5c\u8005\u8bf4\u201cDeepSeek-R1-Zero \u9047\u5230\u4e86\u53ef\u8bfb\u6027\u5dee\u548c\u8bed\u8a00\u6df7\u5408\u7b49\u6311\u6218\u201d\uff0c\u5e76\u5728\u7f51\u4e0a\u5f97\u5230\u4e86\u8bc1\u5b9e\u3002\u7136\u800c\u5728\u6211\u4eec\u7684\u6d4b\u8bd5\u4e2d\uff0c\u6211\u4eec\u53d1\u73b0\u51e0\u4e4e\u6ca1\u6709\u8bc1\u636e\u8868\u660e\u5728 ARC-AGI-1 \u4e0a\u6d4b\u8bd5 R1-Zero \u5b58\u5728\u4e0d\u4e00\u81f4\uff0c\u8fd9\u4e0e\u7cfb\u7edf\u8fdb\u884c RL \u7684\u6570\u5b66\u548c\u7f16\u7801\u9886\u57df\u7c7b\u4f3c\u3002<\/p>\n<p>\u7efc\u4e0a\u6240\u8ff0\uff0c\u8fd9\u4e9b\u53d1\u73b0\u8868\u660e\uff1a<\/p>\n<ul>\n<li>SFT\uff08\u4f8b\u5982\u4eba\u7c7b\u4e13\u5bb6\u6807\u8bb0\uff09\u5bf9\u4e8e\u5728\u5177\u6709\u5f3a\u9a8c\u8bc1\u7684\u9886\u57df\u4e2d\u8fdb\u884c\u51c6\u786e\u548c\u6e05\u6670\u7684 CoT \u63a8\u7406\u4e0d\u662f\u5fc5\u9700\u7684\u3002<\/li>\n<li>R1-Zero \u8bad\u7ec3\u8fc7\u7a0b\u80fd\u591f\u901a\u8fc7 RL \u4f18\u5316\u5728\u6807\u8bb0\u7a7a\u95f4\u4e2d\u521b\u5efa\u81ea\u5df1\u7684\u5185\u90e8\u9886\u57df\u7279\u5b9a\u8bed\u8a00\uff08\u201cDSL\u201d\uff09\u3002<\/li>\n<li>SFT \u5bf9\u4e8e\u63d0\u9ad8 CoT \u63a8\u7406\u9886\u57df\u7684\u901a\u7528\u6027\u662f\u5fc5\u8981\u7684\u3002<\/li>\n<\/ul>\n<p>\u8fd9\u5f88\u76f4\u89c2\uff0c\u56e0\u4e3a\u8bed\u8a00\u672c\u8eab\u5b9e\u9645\u4e0a\u662f\u4e00\u79cd\u63a8\u7406 DSL\u3002 \u5b8c\u5168\u76f8\u540c\u7684\u201c\u5355\u8bcd\u201d\u53ef\u4ee5\u5728\u4e00\u4e2a\u9886\u57df\u4e2d\u5b66\u4e60\u5e76\u5e94\u7528\u4e8e\u53e6\u4e00\u4e2a\u9886\u57df\uff0c\u5c31\u50cf\u4e00\u4e2a\u7a0b\u5e8f\u4e00\u6837\u3002 \u7eaf RL \u65b9\u6cd5\u8fd8\u4e0d\u80fd\u53d1\u73b0\u5e7f\u6cdb\u7684\u5171\u4eab\u8bcd\u6c47\u8868\uff0c\u6211\u9884\u8ba1\u8fd9\u5c06\u6210\u4e3a\u672a\u6765\u7814\u7a76\u7684\u91cd\u70b9\u3002<\/p>\n<p>\u6700\u7ec8\uff0cR1-Zero \u5c55\u793a\u4e86\u4e00\u79cd\u6f5c\u5728\u7684\u6269\u5c55\u673a\u5236\u7684\u539f\u578b\uff0c\u6ca1\u6709\u4efb\u4f55\u4eba\u7c7b\u74f6\u9888\u2014\u2014\u5373\u4f7f\u5728\u8bad\u7ec3\u6570\u636e\u83b7\u53d6\u672c\u8eab\u4e2d\u4e5f\u662f\u5982\u6b64\u3002<\/p>\n<p>\u51e0\u4e4e\u53ef\u4ee5\u80af\u5b9a\uff0cDeepSeek \u5df2\u7ecf\u5c06\u76ee\u5149\u6295\u5411\u4e86 OpenAI \u7684 o3 \u7cfb\u7edf\u3002\u91cd\u8981\u7684\u662f\u89c2\u5bdf SFT \u6700\u7ec8\u662f\u5426\u6210\u4e3a\u6dfb\u52a0 CoT \u641c\u7d22\u548c\u91c7\u6837\u7684\u5fc5\u8981\u6761\u4ef6\uff0c\u6216\u8005\u5047\u8bbe\u7684\u201cR2-Zero\u201d\u662f\u5426\u53ef\u80fd\u5b58\u5728\u4e8e\u76f8\u540c\u7684\u5bf9\u6570\u7cbe\u5ea6\u4e0e\u63a8\u7406\u7f29\u653e\u66f2\u7ebf\u4e0a\u3002\u57fa\u4e8e R1-Zero \u7ed3\u679c\uff0c\u6211\u76f8\u4fe1\u5728\u8fd9\u4e2a\u5047\u8bbe\u7684\u6269\u5c55\u7248\u672c\u4e2d\uff0cSFT \u4e0d\u9700\u8981\u51fb\u8d25 ARC-AGI-1\u3002<\/p>\n<h2>3\u3001\u4e3a\u53ef\u9760\u6027\u800c\u82b1\u94b1<\/h2>\n<p>\u4ece\u7ecf\u6d4e\u89d2\u5ea6\u6765\u770b\uff0c\u4eba\u5de5\u667a\u80fd\u6b63\u5728\u53d1\u751f\u4e24\u5927\u8f6c\u53d8\uff1a<\/p>\n<ul>\n<li>\u73b0\u5728\u53ef\u4ee5\u82b1\u66f4\u591a\u7684\u94b1\u6765\u83b7\u5f97\u66f4\u9ad8\u7684\u51c6\u786e\u6027\u548c\u53ef\u9760\u6027<\/li>\n<li>\u8bad\u7ec3\u8d44\u91d1\u6b63\u5728\u8f6c\u5411\u63a8\u7406\u8d44\u91d1<\/li>\n<\/ul>\n<p>\u4e24\u8005\u90fd\u5c06\u63a8\u52a8\u5bf9\u63a8\u7406\u7684\u5927\u91cf\u9700\u6c42\uff0c\u800c\u4e14\u90fd\u4e0d\u4f1a\u51cf\u5c11\u5bf9\u66f4\u591a\u8ba1\u7b97\u7684\u9700\u6c42\u3002\u4e8b\u5b9e\u4e0a\uff0c\u5b83\u4eec\u4f1a\u589e\u52a0\u5bf9\u8ba1\u7b97\u7684\u9700\u6c42\u3002<\/p>\n<p>\u4eba\u5de5\u667a\u80fd\u63a8\u7406\u7cfb\u7edf\u627f\u8bfa\u7684\u56de\u62a5\u8fdc\u4e0d\u6b62\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u66f4\u9ad8\u7684\u51c6\u786e\u6027\u3002\u963b\u788d\u66f4\u591a\u4eba\u5de5\u667a\u80fd\u81ea\u52a8\u5316\u4f7f\u7528\uff08\u4f8b\u5982\u63a8\u7406\u9700\u6c42\uff09\u7684\u9996\u8981\u95ee\u9898\u662f\u53ef\u9760\u6027\u3002\u6211\u66fe\u4e0e\u6570\u767e\u540d\u8bd5\u56fe\u5728\u5176\u4e1a\u52a1\u4e2d\u90e8\u7f72 AI \u4ee3\u7406\u7684 Zapier \u5ba2\u6237\u4ea4\u8c08\u8fc7\uff0c\u4ed6\u4eec\u7684\u53cd\u9988\u975e\u5e38\u4e00\u81f4\uff1a\u201c\u6211\u8fd8\u4e0d\u4fe1\u4efb\u5b83\u4eec\uff0c\u56e0\u4e3a\u5b83\u4eec\u5de5\u4f5c\u4e0d\u53ef\u9760\u201d\u3002<\/p>\n<p>\u4e4b\u524d\u6211\u66fe\u8ba4\u4e3a\uff0cARC-AGI \u7684\u8fdb\u5c55\u5c06\u5e26\u6765\u66f4\u9ad8\u7684\u53ef\u9760\u6027\u3002LLM \u4ee3\u7406\u9762\u4e34\u7684\u6311\u6218\u662f\uff0c\u5b83\u4eec\u9700\u8981\u5f3a\u5927\u7684\u672c\u5730\u57df\u63a7\u5236\u624d\u80fd\u53ef\u9760\u5730\u5de5\u4f5c\u3002\u66f4\u5f3a\u7684\u6cdb\u5316\u80fd\u529b\u9700\u8981\u80fd\u591f\u9002\u5e94\u672a\u77e5\u60c5\u51b5\u3002\u6211\u4eec\u73b0\u5728\u5f00\u59cb\u770b\u5230\u8bc1\u636e\u8868\u660e\u8fd9\u79cd\u89c2\u70b9\u662f\u6b63\u786e\u7684\u3002\u56e0\u6b64\uff0c\u51e0\u5bb6\u516c\u53f8\u73b0\u5728\u6b63\u5728\u5f15\u5165\u4ee3\u7406\uff08Anthropic\u3001OpenAI\u3001Apple \u7b49\uff09\u4e5f\u5c31\u4e0d\u8db3\u4e3a\u5947\u4e86\u3002<\/p>\n<p>\u7531\u4e8e\u53ef\u9760\u6027\u9700\u6c42\uff0c\u4ee3\u7406\u5c06\u63a8\u52a8\u8fd1\u671f\u9700\u6c42\u63a8\u65ad\u7684\u5927\u5e45\u589e\u957f\u3002\u66f4\u5e7f\u6cdb\u5730\u8bf4\uff0c\u5f00\u53d1\u4eba\u5458\u53ef\u4ee5\u9009\u62e9\u6295\u5165\u66f4\u591a\u8ba1\u7b97\u6765\u589e\u52a0\u7528\u6237\u5bf9\u7cfb\u7edf\u7684\u4fe1\u4efb\u3002\u7136\u800c\uff0c\u66f4\u9ad8\u7684\u53ef\u9760\u6027\u5e76\u4e0d\u610f\u5473\u7740 100% \u7684\u51c6\u786e\u6027\u2014\u2014\u4f46\u4f60\u4f1a\u671f\u671b\u66f4\u4e00\u81f4\u7684\u4e0d\u51c6\u786e\u6027\u3002\u8fd9\u6ca1\u5173\u7cfb\uff0c\u56e0\u4e3a\u7528\u6237\u548c\u5f00\u53d1\u4eba\u5458\u73b0\u5728\u53ef\u4ee5\u5728\u51c6\u786e\u5ea6\u8f83\u4f4e\u65f6\u901a\u8fc7\u63d0\u793a\u66f4\u81ea\u4fe1\u5730\u5f15\u5bfc\u884c\u4e3a\u3002<\/p>\n<p>\u4ee5\u524d\u8ba1\u7b97\u673a\u65e0\u6cd5\u89e3\u51b3\u7684\u95ee\u9898\u73b0\u5728\u9700\u8981\u82b1\u8d39\u5927\u91cf\u91d1\u94b1\u3002\u968f\u7740\u6548\u7387\u7684\u63d0\u9ad8\uff0c\u8fd9\u4e9b\u91d1\u989d\u5c06\u4f1a\u4e0b\u964d\u3002<\/p>\n<h2>4\u3001\u63a8\u7406\u5373\u8bad\u7ec3<\/h2>\n<p>\u53e6\u4e00\u4e2a\u91cd\u5927\u8f6c\u53d8\u662f\u8fdb\u5165 LLM \u7cfb\u7edf\u8fdb\u884c\u9884\u8bad\u7ec3\u7684\u6570\u636e\u6765\u6e90\u3002\u4ee5\u524d\uff0c\u5927\u591a\u6570\u6570\u636e\u90fd\u662f\u4ece\u73b0\u6709\u7684 LLM \u8d2d\u4e70\u3001\u6293\u53d6\u6216\u5408\u6210\u751f\u6210\u7684\uff08\u4f8b\u5982\u63d0\u70bc\u6216\u589e\u5f3a\uff09\u3002<\/p>\n<p>\u8fd9\u4e9b\u63a8\u7406\u7cfb\u7edf\u63d0\u4f9b\u4e86\u4e00\u79cd\u65b0\u9009\u62e9\uff0c\u5373\u751f\u6210\u201c\u771f\u5b9e\u201d\u6570\u636e\u800c\u4e0d\u662f\u201c\u5408\u6210\u201d\u6570\u636e\u3002\u4eba\u5de5\u667a\u80fd\u884c\u4e1a\u4f7f\u7528\u5408\u6210\u4e00\u8bcd\u6765\u8bc6\u522b\u901a\u5e38\u901a\u8fc7 LLM \u56de\u6536\u7684\u4f4e\u8d28\u91cf\u6570\u636e\uff0c\u4ee5\u589e\u52a0\u603b\u4f53\u8bad\u7ec3\u6570\u636e\u91cf &#8211; \u6536\u76ca\u9012\u51cf\u3002<\/p>\n<p>\u4f46\u73b0\u5728\u6709\u4e86\u63a8\u7406\u7cfb\u7edf\u548c\u9a8c\u8bc1\u5668\uff0c\u6211\u4eec\u53ef\u4ee5\u521b\u5efa\u5168\u65b0\u7684\u5408\u6cd5\u6570\u636e\u8fdb\u884c\u8bad\u7ec3\u3002\u8fd9\u53ef\u4ee5\u79bb\u7ebf\u5b8c\u6210\uff0c\u5f00\u53d1\u4eba\u5458\u4ed8\u8d39\u521b\u5efa\u6570\u636e\uff0c\u4e5f\u53ef\u4ee5\u5728\u63a8\u7406\u65f6\u5b8c\u6210\uff0c\u6700\u7ec8\u7528\u6237\u4ed8\u8d39\uff01<\/p>\n<p>\u8fd9\u662f\u7ecf\u6d4e\u5b66\u4e2d\u4e00\u4e2a\u4ee4\u4eba\u7740\u8ff7\u7684\u8f6c\u53d8\uff0c\u8868\u660e\u62e5\u6709\u6700\u591a\u4ed8\u8d39\u5ba2\u6237\u7684\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u5f00\u53d1\u4eba\u5458\u53ef\u80fd\u4f1a\u51fa\u73b0\u6743\u529b\u96c6\u4e2d\u5931\u63a7\u7684\u65f6\u523b\u3002\u8fd9\u4e9b\u5ba2\u6237\u6b63\u5728\u4e3a\u521b\u5efa\u65b0\u7684\u9ad8\u8d28\u91cf\u6570\u636e\u4e70\u5355\u2026\u2026\u8fd9\u6539\u8fdb\u4e86\u6a21\u578b\u2026\u2026\u53d8\u5f97\u66f4\u597d\uff0c\u66f4\u53d7\u7528\u6237\u9752\u7750\u2026\u2026\u4f60\u660e\u767d\u4e86\u3002<\/p>\n<p>\u5982\u679c\u6211\u4eec\u80fd\u591f\u7a81\u7834\u4eba\u7c7b\u4e13\u5bb6 CoT \u969c\u788d\uff0c\u521b\u5efa\u4e00\u4e2a\u6781\u5176\u9ad8\u6548\u7684\u7cfb\u7edf\uff0c\u901a\u8fc7\u641c\u7d22\/\u5408\u6210\u548c\u9a8c\u8bc1\u6765\u521b\u5efa\u65b0\u6570\u636e\uff0c\u90a3\u4e48\u6211\u4eec\u5e94\u8be5\u671f\u5f85\u5927\u91cf\u8ba1\u7b97\u6d8c\u5165\u8fd9\u4e9b\u63a8\u7406\u7cfb\u7edf\uff0c\u56e0\u4e3a\u5b83\u4eec\u786e\u5b9e\u53ea\u9700\u8f93\u5165\u7f8e\u5143\u548c\u539f\u59cb\u6570\u636e\u5c31\u80fd\u53d8\u5f97\u66f4\u597d\u3002\u6700\u7ec8\uff0c\u8fd9\u79cd\u7c7b\u578b\u7684\u4eba\u5de5\u667a\u80fd\u8bad\u7ec3\u5c06\u5b8c\u5168\u8d85\u8d8a\u5bf9\u4eba\u7c7b\u751f\u6210\u7684\u6570\u636e\u7684\u9884\u8bad\u7ec3\u3002<\/p>\n<h2>5\u3001\u7ed3\u675f\u8bed<\/h2>\n<p>\u968f\u7740\u63a8\u7406\u9700\u6c42\u7684\u589e\u52a0\u53d8\u5f97\u6e05\u6670\uff0c\u6211\u4eec\u5c06\u7ee7\u7eed\u770b\u5230\u5e02\u573a\u8c03\u6574\u3002\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u6548\u7387\u53ea\u4f1a\u63a8\u52a8\u66f4\u591a\u7684\u4f7f\u7528\uff0c\u8fd9\u4e0d\u4ec5\u662f\u56e0\u4e3a\u6770\u6587\u65af\u6096\u8bba\uff0c\u8fd8\u56e0\u4e3a\u968f\u7740\u6548\u7387\u7684\u63d0\u9ad8\uff0c\u65b0\u7684\u8bad\u7ec3\u673a\u5236\u88ab\u89e3\u9501\u3002<\/p>\n<p>\u968f\u7740 R1 \u7684\u5f00\u653e\u548c\u53ef\u590d\u5236\uff0c\u66f4\u591a\u7684\u4eba\u548c\u56e2\u961f\u5c06\u628a CoT \u548c\u641c\u7d22\u63a8\u5411\u6781\u9650\u3002\u8fd9\u5c06\u66f4\u5feb\u5730\u544a\u8bc9\u6211\u4eec\u524d\u6cbf\u5728\u54ea\u91cc\uff0c\u5b83\u5c06\u63a8\u52a8\u4e00\u6ce2\u521b\u65b0\u6d6a\u6f6e\uff0c\u4ece\u800c\u589e\u52a0\u5feb\u901f\u5b9e\u73b0 AGI \u7684\u673a\u4f1a\u3002<\/p>\n<p>\u5df2\u7ecf\u6709\u51e0\u4e2a\u4eba\u544a\u8bc9\u6211\uff0c\u4ed6\u4eec\u8ba1\u5212\u5728 2025 \u5e74 ARC \u5956\u4e2d\u4f7f\u7528 R1 \u5f0f\u7cfb\u7edf\uff0c\u6211\u5f88\u9ad8\u5174\u770b\u5230\u7ed3\u679c\u3002<\/p>\n<p>R1 \u5f00\u653e\u8fd9\u4e00\u4e8b\u5b9e\u5bf9\u4e16\u754c\u6765\u8bf4\u662f\u4e00\u4ef6\u5927\u4e8b\u3002 DeepSeek \u63a8\u52a8\u4e86\u79d1\u5b66\u524d\u6cbf\u7684\u53d1\u5c55\u3002<\/p>\n<hr>\n","protected":false},"excerpt":{"rendered":"<p>\u6211\u4eec\u8fd8\u6ca1\u6709 AGI\uff0c\u521b\u65b0\u4ecd\u7136\u53d7\u5230\u9650\u5236\u2014\u2014\u6269\u5927\u7eaf LLM \u9884\u8bad\u7ec3\u5e76\u4e0d\u662f\u51fa\u8def\uff0c\u5c3d\u7ba1\u8fd9\u662f\u53bb\u5e74\u590f\u5929\u4eba\u5de5\u667a\u80fd\u884c\u4e1a\u7684\u4e3b\u6d41\u53d9\u4e8b\u548c\u4e3b\u6d41\u516c\u4f17\u89c2\u70b9\u3002 \u53d9\u4e8b\u4e4b\u6240\u4ee5\u91cd\u8981\uff0c\u662f\u56e0\u4e3a\u5b83\u4eec\u6700\u7ec8\u4f1a\u63a8\u52a8\u7ecf\u6d4e\u6d3b\u52a8\uff0c\u5982\u6295\u8d44\u3001\u7814\u7a76\u91cd\u70b9\u3001\u8d44\u91d1\u3001\u5730\u7f18\u653f\u6cbb\u3001\u8d38\u6613\u7b49\u3002\u4f8b\u5982\uff0c\u5728 2023-24 \u5e74\uff0c\u65b0 LLM \u521d\u521b\u516c\u53f8\u7684\u6295\u8d44\u7ea6\u4e3a 200 \u4ebf\u7f8e\u5143\uff0c\u800c\u65b0 AGI \u521d\u521b\u516c\u53f8\u7684\u6295\u8d44\u4ec5\u4e3a\u7ea6 2 \u4ebf\u7f8e\u5143\u3002 \u6211\u4eec\u4e8e\u53bb\u5e74 6 \u6708\u63a8\u51fa\u4e86 ARC Prize 2024\uff0c\u4ee5\u63d0\u9ad8\u4eba\u4eec\u5bf9 LLM \u6269\u5c55\u9650\u5236\u7684\u8ba4\u8bc6\uff0c\u5e76\u63a8\u52a8\u4e00\u4e2a\u6709\u7528\u7684\u57fa\u51c6 ARC-AGI-1 \u671d\u7740\u65b0\u7684\u65b9\u5411\u53d1\u5c55\uff0c\u5373\u8981\u6c42 AI \u7cfb\u7edf\u9002\u5e94\u65b0\u7684\u3001\u770b\u4e0d\u89c1\u7684\u95ee\u9898\uff0c\u800c\u4e0d\u662f\u4e25\u683c\u4f9d\u8d56\u8bb0\u5fc6\u3002 1\u3001DeepSeek R1\u8bad\u7ec3\u67b6\u6784 R1 \u8bad\u7ec3\u67b6\u6784\u3002 \u4e0a\u5468\uff0cDeepSeek \u53d1\u5e03\u4e86\u4ed6\u4eec\u7684\u65b0 R1-Zero \u548c R1\u201c\u63a8\u7406\u673a\u201d\u7cfb\u7edf\uff0c\u53ef\u4e0e OpenAI \u7684 ARC-AGI-1 \u4e0a\u7684 o1 \u7cfb\u7edf\u76f8\u5ab2\u7f8e\u3002R1-Zero\u3001R1 \u548c o1\uff08\u4f4e\u8ba1\u7b97\uff09\u7684\u5f97\u5206\u90fd\u5728 15-20% \u5de6\u53f3\u2014\u2014\u76f8\u6bd4\u4e4b\u4e0b\uff0cGPT-4o \u7684\u5f97\u5206\u4e3a 5%\uff0c\u8fd9\u662f\u591a\u5e74\u6765\u7eaf LLM \u6269\u5c55\u7684\u5dc5\u5cf0\u3002\u6839\u636e\u672c\u5468\u7f8e\u56fd\u5e02\u573a\u7684\u53cd\u5e94\uff0c\u516c\u4f17\u4e5f\u5f00\u59cb\u4e86\u89e3\u7eaf LLM \u6269\u5c55\u7684\u5c40\u9650\u6027\u3002\u7136\u800c\uff0c\u516c\u4f17\u5bf9\u5373\u5c06\u5230\u6765\u7684\u63a8\u7406\u9700\u6c42\u4ecd\u7136\u666e\u904d\u4e0d\u4e86\u89e3\u3002 2024 \u5e74 12 [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[13],"tags":[],"class_list":["post-53774","post","type-post","status-publish","format-standard","hentry","category-ai"],"_links":{"self":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/posts\/53774","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/comments?post=53774"}],"version-history":[{"count":0,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/posts\/53774\/revisions"}],"wp:attachment":[{"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/media?parent=53774"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/categories?post=53774"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/fwq.ai\/blog\/wp-json\/wp\/v2\/tags?post=53774"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}