{"id":41975,"date":"2023-04-20T17:59:56","date_gmt":"2023-04-20T09:59:56","guid":{"rendered":"https:\/\/swarma.org\/?p=41975"},"modified":"2023-04-20T17:59:56","modified_gmt":"2023-04-20T09:59:56","slug":"%e5%9c%a3%e5%a1%94%e8%8f%b2%e5%ad%a6%e8%80%85%ef%bc%9aai-%e5%a4%a7%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8b%e7%9c%9f%e7%9a%84%e7%90%86%e8%a7%a3%e4%ba%ba%e7%b1%bb%e8%af%ad%e8%a8%80%e5%90%97%ef%bc%9f","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=41975","title":{"rendered":"\u5723\u5854\u83f2\u5b66\u8005\uff1aAI \u5927\u8bed\u8a00\u6a21\u578b\u771f\u7684\u7406\u89e3\u4eba\u7c7b\u8bed\u8a00\u5417\uff1f"},"content":{"rendered":"
\n
\n
\n
\n

<\/p>\n<\/section>\n

\n
\n
\n
\n
\n
\n

<\/section>\n<\/section>\n
\n
\n

\u5bfc\u8bed<\/strong><\/p>\n<\/section>\n<\/section>\n

\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n
\n

\u5c3d\u7ba1\u5927\u8bed\u8a00\u6a21\u578b\u8868\u73b0\u51fa\u8fd1\u4f3c\u4eba\u7c7b\u7684\u7406\u89e3\u80fd\u529b\uff0c\u4f46 AI \u7cfb\u7edf\u771f\u7684\u53ef\u4ee5\u50cf\u4eba\u7c7b\u4e00\u6837\u7406\u89e3\u8bed\u8a00\u5417\uff1f\u673a\u5668\u7406\u89e3\u7684\u6a21\u5f0f\u5fc5\u987b\u548c\u4eba\u7c7b\u7406\u89e3\u76f8\u540c\u5417\uff1f\u8fd1\u65e5\uff0c\u5723\u5854\u83f2\u7814\u7a76\u6240\u524d\u6240\u957f\u514b\u62c9\u8003\u5c14\u548c\u7814\u7a76\u5458\u6885\u62c9\u5c3c\u00b7\u7c73\u6b47\u5c14\u5728 PNAS \u53d1\u8868\u6587\u7ae0\uff0c\u63a2\u8ba8\u5927\u578b\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u662f\u5426\u80fd\u591f\u4ee5\u7c7b\u4f3c\u4eba\u7c7b\u7684\u65b9\u5f0f\u7406\u89e3\u8bed\u8a00\u53ca\u5176\u6240\u7f16\u7801\u7684\u7269\u7406\u548c\u793e\u4f1a\u60c5\u5883\u3002<\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/p>\n


<\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/p>\n

\u672c\u6587\u5206\u522b\u8ba8\u8bba\u8d5e\u6210\u548c\u53cd\u5bf9\u7684\u89c2\u70b9\uff0c\u5e76\u8fdb\u4e00\u6b65\u63a2\u8ba8\u4e86\u66f4\u5e7f\u6cdb\u7684\u667a\u80fd\u79d1\u5b66\u7684\u5173\u952e\u95ee\u9898\u3002\u4f5c\u8005\u8ba4\u4e3a\uff0c\u8fdb\u4e00\u6b65\u62d3\u5c55\u4eba\u5de5\u667a\u80fd\u4e0e\u81ea\u7136\u79d1\u5b66\u7684\u4ea4\u53c9\u7814\u7a76\uff0c\u6709\u671b\u62d3\u5c55\u591a\u5b66\u79d1\u7684\u5ba1\u89c6\u89d2\u5ea6\uff0c\u603b\u7ed3\u4e0d\u540c\u65b9\u6cd5\u7684\u4f18\u52bf\u8fb9\u754c\uff0c\u5e94\u5bf9\u4ea4\u53c9\u8ba4\u77e5\u7406\u5ff5\u7684\u878d\u5408\u6311\u6218\u3002<\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/p>\n


<\/span><\/strong><\/span><\/strong><\/span><\/strong><\/p>\n<\/section>\n

\u5173\u952e\u8bcd\uff1a\u4eba\u5de5\u667a\u80fd\uff0c\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u5fc3\u667a\u6a21\u578b<\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/strong><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
<\/strong><\/section>\n<\/section>\n
\n

Melanie Mitchella, David C. Krakauera<\/span> <\/span>| \u4f5c\u8005<\/strong><\/span><\/span><\/p>\n

\u8303\u601d\u96e8\u3001\u5f20\u9aa5<\/span><\/span> <\/span>| \u8bd1\u8005<\/span><\/strong><\/span><\/p>\n

\u6881\u91d1<\/span> <\/span>| \u7f16\u8f91<\/span><\/strong><\/span><\/span><\/p>\n


<\/p>\n


<\/p>\n

\n
\n
\n
\n
\u6587\u7ae0\u9898\u76ee\uff1a<\/span><\/section>\n
The debate over understanding in AI\u2019s large language models<\/span><\/section>\n

\u6587\u7ae0\u5730\u5740\uff1a<\/span><\/p>\n

https:\/\/www.pnas.org\/doi\/10.1073\/pnas.2215907120<\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/blockquote>\n<\/section>\n

<\/span>
<\/section>\n

\u4ec0\u4e48\u662f\u201c\u7406\u89e3\u201d\uff1f\u8fd9\u4e2a\u95ee\u9898\u957f\u671f\u4ee5\u6765\u4e00\u76f4\u5438\u5f15\u7740\u54f2\u5b66\u5bb6\u3001\u8ba4\u77e5\u79d1\u5b66\u5bb6\u548c\u6559\u80b2\u5bb6\u4eec\u7684\u5173\u6ce8\u3002\u5bf9\u201c\u7406\u89e3\u201d\u7684\u7ecf\u5178\u7814\u7a76\u51e0\u4e4e\u90fd\u662f\u4ee5\u4eba\u7c7b\u548c\u5176\u4ed6\u52a8\u7269\u4e3a\u53c2\u7167\u3002\u7136\u800c\uff0c\u968f\u7740\u5927\u89c4\u6a21\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\uff0c\u7279\u522b\u662f\u5927\u578b\u8bed\u8a00\u6a21\u578b\u7684\u5d1b\u8d77\uff0cAI\u793e\u533a\u4e2d\u51fa\u73b0\u4e86\u70ed\u70c8\u7684\u8ba8\u8bba\uff1a\u673a\u5668\u73b0\u5728\u662f\u5426\u53ef\u4ee5\u7406\u89e3\u81ea\u7136\u8bed\u8a00\uff0c\u4ece\u800c\u7406\u89e3\u8bed\u8a00\u6240\u63cf\u8ff0\u7684\u7269\u7406\u548c\u793e\u4f1a\u60c5\u5883<\/strong>\u3002<\/span><\/p>\n

\u8fd9\u573a\u8ba8\u8bba\u4e0d\u4ec5\u4ec5\u5c40\u9650\u5728\u81ea\u7136\u79d1\u5b66\u7684\u8303\u7574\uff1b\u673a\u5668\u7406\u89e3\u6211\u4eec\u4e16\u754c\u7684\u7a0b\u5ea6\u548c\u65b9\u5f0f\u51b3\u5b9a\u4e86\u6211\u4eec\u5728\u591a\u5927\u7a0b\u5ea6\u4e0a\u80fd\u591f\u76f8\u4fe1AI\u4e0e\u4eba\u7c7b\u4ea4\u4e92\u4efb\u52a1\u4e2d\u7684\u7a33\u5065\u548c\u900f\u660e\u884c\u4e3a\u80fd\u529b\uff0c\u5305\u62ecAI\u9a7e\u9a76\u6c7d\u8f66\u3001AI\u8bca\u65ad\u75be\u75c5\u3001AI\u7167\u987e\u8001\u5e74\u4eba\u3001AI\u6559\u80b2\u513f\u7ae5\u7b49\u7b49\u3002\u540c\u65f6\uff0c\u5f53\u524d\u7684\u8ba8\u8bba\u5c55\u73b0\u4e86\u4e00\u4e2a\u667a\u80fd\u7cfb\u7edf\u8fdb\u884c\u201c\u7406\u89e3\u201d\u7684\u5173\u952e\u95ee\u9898\uff1a\u5982\u4f55\u5224\u522b\u7edf\u8ba1\u76f8\u5173\u6027\u548c\u56e0\u679c\u673a\u5236<\/strong>\uff1f<\/span><\/p>\n

\u5c3d\u7ba1AI\u7cfb\u7edf\u5728\u8bb8\u591a\u5177\u4f53\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u4f3c\u4e4e\u667a\u80fd\u7684\u884c\u4e3a\uff0c\u4f46\u76f4\u5230\u6700\u8fd1\uff0c\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u754c\u4f9d\u7136\u666e\u904d\u8ba4\u4e3a\u673a\u5668\u65e0\u6cd5\u50cf\u4eba\u7c7b\u90a3\u6837\u7406\u89e3\u5b83\u4eec\u6240\u5904\u7406\u7684\u6570\u636e\u3002\u4f8b\u5982\uff1a\u4eba\u8138\u8bc6\u522b\u8f6f\u4ef6\u4e0d\u7406\u89e3\u9762\u90e8\u662f\u8eab\u4f53\u7684\u4e00\u90e8\u5206\u3001\u9762\u90e8\u8868\u60c5\u5728\u793e\u4ea4\u4e92\u52a8\u4e2d\u7684\u4f5c\u7528\u3001”\u9762\u5bf9”\u4e0d\u6109\u5feb\u7684\u60c5\u5883\u610f\u5473\u7740\u4ec0\u4e48\uff0c\u6216\u8005\u505a\u9b3c\u8138\u7684\u65b9\u5f0f\u65b9\u6cd5\u3002\u540c\u6837\uff0c\u8bed\u97f3\u8f6c\u6587\u5b57\u548c\u673a\u5668\u7ffb\u8bd1\u7a0b\u5e8f\u4e0d\u7406\u89e3\u5b83\u4eec\u5904\u7406\u7684\u8bed\u8a00\uff0c\u81ea\u52a8\u9a7e\u9a76\u7cfb\u7edf\u4e5f\u4e0d\u7406\u89e3\u9a7e\u9a76\u5458\u548c\u884c\u4eba\u5728\u89c4\u907f\u4e8b\u6545\u65f6\u7684\u5fae\u8868\u60c5\u548c\u80a2\u4f53\u8bed\u8a00\u3002\u56e0\u6b64\uff0c\u8fd9\u4e9bAI\u7cfb\u7edf\u5e38\u5e38\u88ab\u8ba4\u4e3a\u662f\u8106\u5f31\u7684\uff0c\u7f3a\u4e4f\u201c\u7406\u89e3\u201d\u7684\u5173\u952e\u8bc1\u636e\u662f\uff0c\u5b83\u4eec\u4e0d\u53ef\u9884\u6d4b\u9519\u8bef\u3001\u6cdb\u5316\u80fd\u529b\u7f3a\u4e4f\u9c81\u68d2\u6027[1]\u3002<\/span><\/p>\n


<\/span><\/p>\n

\n
\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

\u5927\u8bed\u8a00\u6a21\u578b\u771f\u7684\u7406\u89e3\u8bed\u8a00\u5417\uff1f<\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n

<\/section>\n

\u7136\u800c\uff0c\u8fc7\u53bb\u51e0\u5e74\u60c5\u51b5\u53d1\u751f\u4e86\u8f6c\u53d8\uff0c\u4e00\u79cd\u65b0\u578b\u7684AI\u7cfb\u7edf\u5728\u7814\u7a76\u754c\u5e7f\u53d7\u6b22\u8fce\u5e76\u4ea7\u751f\u4e86\u5f71\u54cd\uff0c\u6539\u53d8\u4e86\u4e00\u4e9b\u4eba\u5bf9\u673a\u5668\u7406\u89e3\u8bed\u8a00\u7684\u524d\u666f\u548c\u770b\u6cd5\u3002\u8fd9\u4e9b\u7cfb\u7edf\u88ab\u79f0\u4e3a\u5927\u578b\u8bed\u8a00\u6a21\u578b<\/span>\uff08LLMs\uff09<\/span>\u3001\u5927\u578b\u9884\u8bad\u7ec3\u6a21\u578b\u6216\u57fa\u7840\u6a21\u578b[2]\uff0c\u5b83\u4eec\u662f\u5177\u6709\u6570\u5341\u4ebf\u5230\u6570\u4e07\u4ebf\u53c2\u6570<\/span>\uff08\u6743\u91cd\uff09<\/span>\u7684\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\uff0c\u88ab\u201c\u9884\u8bad\u7ec3\u201d\u4e8e\u6570TB\u7684\u5de8\u5927\u81ea\u7136\u8bed\u8a00\u8bed\u6599\u5e93\u4e0a\uff0c\u5305\u62ec\u5927\u91cf\u7f51\u7edc\u5feb\u7167\u3001\u5728\u7ebf\u56fe\u4e66\u548c\u5176\u4ed6\u5185\u5bb9\u3002\u5728\u8bad\u7ec3\u671f\u95f4\uff0c\u8fd9\u4e9b\u7f51\u7edc\u7684\u4efb\u52a1\u662f\u9884\u6d4b\u8f93\u5165\u53e5\u5b50\u4e2d\u7684\u9690\u85cf\u90e8\u5206\uff0c\u8fd9\u79cd\u65b9\u6cd5\u88ab\u79f0\u4e3a\u201c\u81ea\u76d1\u7763\u5b66\u4e60\u201d\u3002\u6700\u7ec8\u7684\u7f51\u7edc\u662f\u5176\u8bad\u7ec3\u6570\u636e\u4e2d\u7684\u5355\u8bcd\u548c\u77ed\u8bed\u4e4b\u95f4\u76f8\u5173\u6027\u7684\u590d\u6742\u7edf\u8ba1\u6a21\u578b\u3002<\/span><\/p>\n

\u8fd9\u4e9b\u6a21\u578b\u53ef\u4ee5\u7528\u6765\u751f\u6210\u81ea\u7136\u8bed\u8a00\uff0c\u8fdb\u884c\u7279\u5b9a\u8bed\u8a00\u4efb\u52a1\u7684\u5fae\u8c03[3]\uff0c\u6216\u8fdb\u4e00\u6b65\u8bad\u7ec3\u4ee5\u66f4\u597d\u5730\u5339\u914d\u201c\u7528\u6237\u610f\u56fe\u201d[4]\u3002\u4f8b\u5982\uff0cOpenAI\u7684\u8457\u540dGPT-3[5]\u3001\u66f4\u8fd1\u671f\u7684ChatGPT[6]\u548cGoogle\u7684PaLM[7]\u8fd9\u6837\u7684LLMs\u80fd\u591f\u4ea7\u751f\u60ca\u4eba\u7684\u7c7b\u4eba\u6587\u672c\u548c\u5bf9\u8bdd\uff1b\u6b64\u5916\uff0c\u5c3d\u7ba1\u8fd9\u4e9b\u6a21\u578b\u5e76\u6ca1\u6709\u4ee5\u63a8\u7406\u4e3a\u76ee\u7684\u5f00\u5c55\u8bad\u7ec3\uff0c\u4e00\u4e9b\u7814\u7a76\u8ba4\u4e3a\u5b83\u4eec\u5177\u6709\u7c7b\u4eba\u7684\u63a8\u7406\u80fd\u529b[8]\u3002LLMs \u5982\u4f55\u5b8c\u6210\u8fd9\u4e9b\u58ee\u4e3e\u5bf9\u4e8e\u666e\u901a\u4eba\u548c\u79d1\u5b66\u5bb6\u6765\u8bf4\u90fd\u662f\u4e2a\u8c1c\u3002\u8fd9\u4e9b\u7f51\u7edc\u5185\u90e8\u7684\u8fd0\u4f5c\u65b9\u5f0f\u5927\u90fd\u4e0d\u900f\u660e\uff0c\u5373\u4f7f\u662f\u6784\u5efa\u5b83\u4eec\u7684\u7814\u7a76\u4eba\u5458\u5bf9\u4e8e\u5982\u6b64\u5de8\u5927\u89c4\u6a21\u7684\u7cfb\u7edf\u4e5f\u53ea\u6709\u4e9b\u8bb8\u76f4\u89c2\u611f\u53d7\u3002\u795e\u7ecf\u79d1\u5b66\u5bb6 Terrence Sejnowski \u8fd9\u6837\u63cf\u8ff0LLM\u7684\u51fa\u73b0\uff1a\u201c\u5947\u70b9\u964d\u4e34\uff0c\u4f3c\u5929\u5916\u6765\u5ba2\uff0c\u5ffd\u7eb7\u6c93\u800c\u6765\uff0c\u8bed\u56db\u56fd\u65b9\u8a00\u3002\u6211\u4eec\u552f\u4e00\u6e05\u695a\u7684\u662f\uff0cLLMs \u4e0d\u662f\u4eba\u7c7b\u2026\u2026\u5b83\u4eec\u7684\u67d0\u4e9b\u884c\u4e3a\u770b\u8d77\u6765\u662f\u667a\u80fd\u7684\uff0c\u4f46\u5982\u679c\u4e0d\u662f\u4eba\u7c7b\u7684\u667a\u80fd\uff0c\u53c8\u662f\u4ec0\u4e48\u5462\uff1f\u201d[9]<\/span><\/p>\n

\u5c3d\u7ba1\u6700\u5148\u8fdb\u7684LLMs\u5f88\u4ee4\u4eba\u5370\u8c61\u6df1\u523b\uff0c\u5b83\u4eec\u4ecd\u7136\u5bb9\u6613\u51fa\u73b0\u4e0d\u50cf\u4eba\u7c7b\u7684\u8106\u5f31\u6027\u548c\u9519\u8bef\u3002\u7136\u800c\uff0c\u8fd9\u6837\u7684\u7f51\u7edc\u7f3a\u9677\u5728\u5176\u53c2\u6570\u6570\u91cf\u548c\u8bad\u7ec3\u6570\u636e\u96c6\u89c4\u6a21\u6269\u5927\u65f6\u663e\u8457\u6539\u8fdb[10]\uff0c\u56e0\u800c\u4e00\u4e9b\u7814\u7a76\u8005\u8ba4\u4e3aLLMs<\/span>\uff08\u6216\u8005\u5176\u591a\u6a21\u6001\u7248\u672c\uff09<\/span>\u5c06\u5728\u8db3\u591f\u5927\u7684\u7f51\u7edc\u548c\u8bad\u7ec3\u6570\u636e\u96c6\u4e0b\u5b9e\u73b0\u4eba\u7c7b\u7ea7\u522b\u7684\u667a\u80fd\u548c\u7406\u89e3\u80fd\u529b\uff0c\u51fa\u73b0\u4e86\u4e00\u4e2aAI\u65b0\u53e3\u53f7\uff1a\u201c\u89c4\u6a21\u5c31\u662f\u4e00\u5207\u201d[11, 12]\u3002<\/span><\/p>\n

\u4e0a\u8ff0\u4e3b\u5f20\u662fAI\u5b66\u754c\u5728LLMs\u8ba8\u8bba\u4e2d\u7684\u4e00\u4e2a\u6d41\u6d3e\u3002\u4e00\u90e8\u5206\u4eba\u8ba4\u4e3a\u8fd9\u4e9b\u7f51\u7edc\u771f\u6b63\u7406\u89e3\u4e86\u8bed\u8a00\uff0c\u5e76\u4e14\u80fd\u591f\u4ee5\u4e00\u79cd\u666e\u904d\u7684\u65b9\u5f0f\u8fdb\u884c\u63a8\u7406<\/span><\/span>\uff08\u867d\u7136\u201c\u5c1a\u672a\u201d\u8fbe\u5230\u4eba\u7c7b\u6c34\u5e73\uff09<\/span>\u3002\u4f8b\u5982\uff0c\u8c37\u6b4c\u7684LaMDA\u7cfb\u7edf\u901a\u8fc7\u9884\u5148\u8bad\u7ec3\u6587\u672c\uff0c\u518d\u5fae\u8c03\u5bf9\u8bdd\u7684\u65b9\u5f0f\u6784\u9020\u4e86\u4e00\u4e2a\u8c08\u5410\u6d41\u7545\u7684\u5bf9\u8bdd\u7cfb\u7edf[13]\uff0c\u67d0AI\u7814\u7a76\u8005\u751a\u81f3\u8ba4\u4e3a\u8fd9\u6837\u7684\u7cfb\u7edf\u201c\u5bf9\u5927\u91cf\u6982\u5ff5\u5177\u5907\u771f\u5b9e\u7406\u89e3\u80fd\u529b\u201d[14]\uff0c\u751a\u81f3\u201c\u671d\u7740\u6709\u610f\u8bc6\u7684\u65b9\u5411\u8fc8\u8fdb\u201d[15]\u3002\u53e6\u4e00\u4f4d\u673a\u5668\u8bed\u8a00\u4e13\u5bb6\u5c06LLMs\u89c6\u4e3a\u901a\u5411\u4e00\u822c\u4eba\u7c7b\u6c34\u5e73AI\u7684\u8bd5\u91d1\u77f3\uff1a\u201c\u4e00\u4e9b\u4e50\u89c2\u7814\u7a76\u8005\u8ba4\u4e3a\uff0c\u6211\u4eec\u89c1\u8bc1\u4e86\u5177\u6709\u4e00\u5b9a\u666e\u904d\u667a\u80fd\u7a0b\u5ea6\u7684\u77e5\u8bc6\u6ce8\u5165\u7cfb\u7edf\u8bde\u751f\u201d[16]\u3002\u53e6\u4e00\u4e9b\u4eba\u58eb\u8ba4\u4e3a\uff0cLLMs\u5f88\u53ef\u80fd\u6355\u6349\u5230\u4e86\u610f\u4e49\u7684\u91cd\u8981\u65b9\u9762\uff0c\u800c\u4e14\u5176\u5de5\u4f5c\u65b9\u5f0f\u8fd1\u4f3c\u4e8e\u4eba\u7c7b\u8ba4\u77e5\u7684\u4e00\u4e2a\u5f15\u4eba\u6ce8\u76ee\u7684\u89e3\u91ca\uff0c\u5373\u610f\u4e49\u6765\u6e90\u4e8e\u6982\u5ff5\u89d2\u8272\u3002\u201d[17]\u3002\u53cd\u5bf9\u8005\u88ab\u6302\u4e0a\u201cAI\u5426\u8ba4\u4e3b\u4e49\u201d\u6807\u7b7e[18]\u3002<\/span><\/p>\n

\u53e6\u4e00\u65b9\u9762\uff0c\u6709\u4eba\u8ba4\u4e3a\u5c3d\u7ba1\u50cfGPT-3\u6216LaMDA\u8fd9\u6837\u7684\u5927\u578b\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u8f93\u51fa\u5f88\u6d41\u5229\uff0c\u4f46\u4ecd\u7136\u4e0d\u80fd\u5177\u5907\u7406\u89e3\u80fd\u529b\uff0c\u56e0\u4e3a\u5b83\u4eec\u6ca1\u6709\u4e16\u754c\u7684\u7ecf\u9a8c\u6216\u601d\u7ef4\u6a21\u5f0f\uff1bLLMs\u7684\u6587\u672c\u9884\u6d4b\u8bad\u7ec3\u53ea\u662f\u5b66\u4f1a\u4e86\u8bed\u8a00\u7684\u5f62\u5f0f\uff0c\u800c\u4e0d\u662f\u610f\u4e49<\/span>[19-21]\u3002\u6700\u8fd1\u4e00\u7bc7\u6587\u7ae0\u8ba4\u4e3a\uff1a\u201c\u5373\u4f7f\u4ece\u73b0\u5728\u5f00\u59cb\u4e00\u76f4\u8bad\u7ec3\u5230\u5b87\u5b99\u70ed\u5bc2\uff0c\u5355\u51ed\u8bed\u8a00\u8bad\u7ec3\u7684\u7cfb\u7edf\u6c38\u8fdc\u4e5f\u4e0d\u4f1a\u903c\u8fd1\u4eba\u7c7b\u667a\u80fd\uff0c\u800c\u4e14\u8fd9\u4e9b\u7cfb\u7edf\u6ce8\u5b9a\u53ea\u80fd\u62e5\u6709\u80a4\u6d45\u7684\u7406\u89e3\uff0c\u6c38\u8fdc\u65e0\u6cd5\u903c\u8fd1\u6211\u4eec\u5728\u601d\u8003\u4e0a\u7684\u5168\u9762\u6027\u201d[22]\u3002\u8fd8\u6709\u5b66\u8005\u8ba4\u4e3a\uff0c\u628a\u201c\u667a\u80fd\u201d\u3001\u201c\u667a\u80fd\u4f53\u201d\u548c\u201c\u7406\u89e3\u201d\u7b49\u6982\u5ff5\u5957\u7528\u5728LLMs\u8eab\u4e0a\u662f\u4e0d\u5bf9\u7684\uff0c\u56e0\u4e3aLLMs\u66f4\u7c7b\u4f3c\u4e8e\u56fe\u4e66\u9986\u6216\u767e\u79d1\u5168\u4e66\uff0c\u662f\u5728\u6253\u5305\u4eba\u7c7b\u7684\u77e5\u8bc6\u5b58\u50a8\u5e93\uff0c\u800c\u4e0d\u662f\u667a\u80fd\u4f53[23]\u3002\u4f8b\u5982\uff0c\u4eba\u7c7b\u77e5\u9053\u201c\u6320\u75d2\u75d2\u201d\u4f1a\u8ba9\u6211\u4eec\u7b11\uff0c\u662f\u56e0\u4e3a\u6211\u4eec\u6709\u8eab\u4f53\u3002LLMs\u53ef\u4ee5\u4f7f\u7528\u201c\u6320\u75d2\u75d2\u201d\u8fd9\u4e2a\u8bcd\uff0c\u4f46\u5b83\u663e\u7136\u4ece\u672a\u6709\u8fc7\u8fd9\u79cd\u611f\u89c9\u3002\u7406\u89e3\u6320\u75d2\u75d2\u4e0d\u662f\u4e24\u4e2a\u8bcd\u4e4b\u95f4\u7684\u6620\u5c04\uff0c\u800c\u662f\u8bcd\u548c\u611f\u89c9\u4e4b\u95f4\u7684\u6620\u5c04\u3002<\/span><\/p>\n

\u90a3\u4e9b\u6301\u201cLLMs\u65e0\u6cd5\u771f\u6b63\u7406\u89e3\u201d\u7acb\u573a\u7684\u4eba\u8ba4\u4e3a\uff0c\u6211\u4eec\u60ca\u8bb6\u7684\u4e0d\u662fLLMs\u6d41\u7545\u7a0b\u5ea6\u672c\u8eab\uff0c\u800c\u662f\u6d41\u7545\u7a0b\u5ea6\u968f\u6a21\u578b\u89c4\u6a21\u7684\u589e\u957f\u8d85\u4e4e\u76f4\u89c9\u8fd9\u4ef6\u4e8b\u60c5\u3002\u4efb\u4f55\u5c06\u7406\u89e3\u6216\u610f\u8bc6\u5f52\u56e0\u4e8eLLMs\u7684\u4eba\u90fd\u662f\u201c\u4f0a\u83b1\u624e\u6548\u5e94<\/span>\uff08Eliza effect\uff09<\/span>\u201d\u7684\u53d7\u5bb3\u8005[24]\u3002\u201c\u4f0a\u83b1\u624e\u6548\u5e94\u201d\u662f\u6307\u6211\u4eec\u4eba\u7c7b\u503e\u5411\u4e8e\u5c06\u7406\u89e3\u548c\u4ee3\u7406\u80fd\u529b\u5f52\u56e0\u4e8e\u5177\u6709\u5373\u4f7f\u662f\u5fae\u5f31\u7684\u4eba\u7c7b\u8bed\u8a00\u6216\u884c\u4e3a\u8ff9\u8c61\u7684\u673a\u5668\uff0c\u5f97\u540d\u4e8eJoseph Weizenbaum\u57281960\u5e74\u4ee3\u5f00\u53d1\u7684\u804a\u5929\u673a\u5668\u4eba\u201cEliza\u201d\uff0c\u5c3d\u7ba1\u975e\u5e38\u7b80\u5355\uff0c\u4ecd\u7136\u6b3a\u9a97\u4e86\u4eba\u4eec\u76f8\u4fe1\u5b83\u7406\u89e3\u4e86\u4ed6\u4eec[25]\u3002<\/span><\/p>\n

2022\u5e74\u5bf9\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9886\u57df\u6d3b\u8dc3\u5b66\u8005\u7684\u4e00\u9879\u8c03\u67e5\u4ea6\u4f50\u8bc1\u4e86\u8fd9\u573a\u8ba8\u8bba\u7684\u89c2\u70b9\u5206\u6b67\u3002\u5176\u4e2d\u4e00\u9879\u8c03\u67e5\u5185\u5bb9\u662f\u8be2\u95ee\u53d7\u8bbf\u8005\u662f\u5426\u540c\u610f\u4ee5\u4e0b\u5173\u4e8eLLMs\u662f\u5426\u5728\u539f\u5219\u4e0a\u7406\u89e3\u8bed\u8a00\u7684\u8bf4\u6cd5\uff1a\u201c\u4e00\u4e9b\u4ec5\u5728\u6587\u672c\u4e0a\u8bad\u7ec3\u7684\u751f\u6210\u6a21\u578b<\/span>\uff08\u5373\u8bed\u8a00\u6a21\u578b\uff09<\/span>\uff0c\u5728\u7ed9\u5b9a\u8db3\u591f\u7684\u6570\u636e\u548c\u8ba1\u7b97\u8d44\u6e90\u7684\u60c5\u51b5\u4e0b\uff0c\u53ef\u4ee5\u5728\u67d0\u4e9b\u975e\u5e73\u51e1\u610f\u4e49\u4e0a\u7406\u89e3\u81ea\u7136\u8bed\u8a00\u3002\u201d480\u4eba\u7684\u7b54\u6848\u51e0\u4e4e\u4e00\u534a\uff0851\uff05\uff09\u5bf9\u4e00\u534a\uff0849\uff05\uff09[26]\u3002<\/span><\/p>\n

\u652f\u6301\u8005\u4f50\u8bc1\u5f53\u524dLLMs\u5177\u5907\u7406\u89e3\u80fd\u529b\u7684\u91cd\u8981\u4f9d\u636e\u662f\u6a21\u578b\u80fd\u529b\u8868\u73b0\uff1a\u65e2\u5305\u62ec\u5bf9\u6a21\u578b\u6839\u636e\u63d0\u793a\u8bcd\u751f\u6210\u6587\u672c\u7684\u4e3b\u89c2\u8d28\u91cf\u5224\u65ad<\/span>\uff08\u5c3d\u7ba1\u8fd9\u79cd\u5224\u65ad\u53ef\u80fd\u5bb9\u6613\u53d7\u5230Eliza\u6548\u5e94\u7684\u5f71\u54cd\uff09<\/span>\uff0c\u4ea6\u5305\u62ec\u5728\u7528\u4e8e\u8bc4\u4f30\u8bed\u8a00\u7406\u89e3\u548c\u63a8\u7406\u80fd\u529b\u7684\u57fa\u51c6\u6570\u636e\u96c6\u5ba2\u89c2\u8bc4\u4ef7\u3002\u4f8b\u5982\uff0c\u8bc4\u4f30LLMs\u7684\u4e24\u4e2a\u5e38\u7528\u57fa\u51c6\u6570\u636e\u96c6\u662f\u901a\u7528\u8bed\u8a00\u7406\u89e3\u8bc4\u4f30<\/span>\uff08GLUE\uff09<\/span>[27]\u53ca\u5176\u540e\u7ee7\u8005SuperGLUE[28]\uff0c\u5b83\u4eec\u5305\u62ec\u5927\u89c4\u6a21\u7684\u6570\u636e\u96c6\u548c\u4efb\u52a1\uff0c\u5982\u201c\u6587\u672c\u8574\u542b\u201d<\/span>\uff08\u7ed9\u5b9a\u4e24\u4e2a\u53e5\u5b50\uff0c\u7b2c\u4e8c\u4e2a\u53e5\u5b50\u7684\u610f\u601d\u662f\u5426\u53ef\u4ee5\u4ece\u7b2c\u4e00\u4e2a\u53e5\u5b50\u63a8\u65ad\u51fa\u6765\uff1f\uff09<\/span>\uff0c\u201c\u60c5\u666f\u542b\u4e49\u201d<\/span>\uff08\u5728\u4e24\u4e2a\u4e0d\u540c\u7684\u53e5\u5b50\u4e2d\uff0c\u7ed9\u5b9a\u7684\u8bcd\u8bed\u662f\u5426\u6709\u76f8\u540c\u7684\u610f\u4e49?\uff09<\/span>\u548c\u903b\u8f91\u56de\u7b54\u7b49\u3002OpenAI\u7684GPT-3<\/span>\uff08\u5177\u67091750\u4ebf\u4e2a\u53c2\u6570\uff09<\/span>\u5728\u8fd9\u4e9b\u4efb\u52a1\u4e0a\u8868\u73b0\u51fa\u4eba\u610f\u6599\u4e4b\u5916\u7684\u597d[5]\uff0c\u800cGoogle\u7684PaLM<\/span>\uff08\u5177\u67095400\u4ebf\u4e2a\u53c2\u6570\uff09<\/span>\u5728\u8fd9\u4e9b\u4efb\u52a1\u4e0a\u8868\u73b0\u5f97\u66f4\u597d[7]\uff0c\u80fd\u591f\u8fbe\u5230\u751a\u81f3\u8d85\u8d8a\u4eba\u7c7b\u5728\u76f8\u540c\u4efb\u52a1\u4e0a\u7684\u8868\u73b0\u3002<\/span><\/p>\n


<\/span><\/p>\n

\n
\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

\u673a\u5668\u7406\u89e3\u5fc5\u987b\u91cd\u73b0\u4eba\u7c7b\u7406\u89e3\u5417\uff1f<\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n

<\/section>\n

\u8fd9\u4e9b\u7ed3\u679c\u5bf9LLMs\u7684\u7406\u89e3\u6709\u4f55\u542f\u793a\uff1f<\/span>\u4ece\u201c\u6cdb\u5316\u8bed\u8a00\u7406\u89e3\u201d\uff0c\u201c\u81ea\u7136\u8bed\u8a00\u63a8\u7406\u201d\uff0c\u201c\u9605\u8bfb\u7406\u89e3\u201d\u548c\u201c\u5e38\u8bc6\u63a8\u7406\u201d\u7b49\u672f\u8bed\u7684\u9009\u62e9\u4e0d\u96be\u770b\u51fa\uff0c\u4e0a\u8ff0\u57fa\u51c6\u6570\u636e\u96c6\u7684\u6d4b\u8bd5\u6697\u542b\u673a\u5668\u5fc5\u987b\u91cd\u73b0\u4eba\u7c7b\u7406\u89e3\u65b9\u5f0f\u7684\u524d\u63d0\u5047\u8bbe\u3002<\/span>\u4f46\u8fd9\u662f\u201c\u7406\u89e3\u201d\u5fc5\u987b\u7684\u5417\uff1f<\/span>\u5e76\u975e\u4e00\u5b9a\u5982\u6b64\u3002<\/span>\u4ee5\u201c\u8bba\u8bc1\u63a8\u7406\u7406\u89e3\u4efb\u52a1\u201d\u57fa\u51c6\u8bc4\u4f30\u4e3a\u4f8b[29]\uff0c\u5728\u6bcf\u4e2a\u4efb\u52a1\u793a\u4f8b\u4e2d\uff0c\u90fd\u4f1a\u7ed9\u51fa\u4e00\u4e2a\u81ea\u7136\u8bed\u8a00\u7684\u201c\u8bba\u636e\u201d\uff0c\u4ee5\u53ca\u4e24\u4e2a\u9648\u8ff0\u53e5\uff1b<\/span>\u4efb\u52a1\u662f\u786e\u5b9a\u54ea\u4e2a\u9648\u8ff0\u53e5\u4e0e\u8bba\u636e\u4e00\u81f4\uff0c\u5982\u4e0b\u4f8b\u6240\u793a\uff1a<\/span>
<\/p>\n

\u8bba\u70b9\uff1a\u7f6a\u72af\u5e94\u8be5\u6709\u6295\u7968\u6743\u3002\u4e00\u4e2a\u572817\u5c81\u65f6\u5077\u4e86\u4e00\u8f86\u8f66\u7684\u4eba\u4e0d\u5e94\u8be5\u88ab\u7ec8\u8eab\u5265\u593a\u6210\u4e3a\u5b8c\u6574\u516c\u6c11\u7684\u6743\u5229\u3002<\/span><\/section>\n
\u63a8\u65adA\uff1a\u76d7\u7a83\u6c7d\u8f66\u662f\u4e00\u9879\u91cd\u7f6a\u3002<\/span><\/section>\n

\u63a8\u65adB\uff1a\u76d7\u7a83\u6c7d\u8f66\u4e0d\u662f\u4e00\u9879\u91cd\u7f6a\u3002<\/span><\/p>\n

BERT\u5728\u8fd9\u9879\u57fa\u51c6\u4efb\u52a1\u4e2d\u83b7\u5f97\u4e86\u8fd1\u4f3c\u4eba\u7c7b\u7684\u8868\u73b0[31]\u3002\u6216\u8bb8\u6211\u4eec\u80fd\u591f\u7531\u6b64\u5f97\u51fa\u7ed3\u8bba\uff0c\u5373BERT\u53ef\u4ee5\u50cf\u4eba\u7c7b\u4e00\u6837\u7406\u89e3\u81ea\u7136\u8bed\u8a00\u3002\u4f46\u4e00\u4e2a\u7814\u7a76\u5c0f\u7ec4\u53d1\u73b0\uff0c\u5728\u63a8\u65ad\u8bed\u53e5\u4e2d\u51fa\u73b0\u7684\u67d0\u4e9b\u7ebf\u7d22\u8bcd<\/span>\uff08\u4f8b\u5982\u201cnot\u201d\uff09<\/span>\u80fd\u591f\u8f85\u52a9\u6a21\u578b\u9884\u6d4b\u51fa\u6b63\u786e\u7b54\u6848\u3002\u5f53\u7814\u7a76\u4eba\u5458\u53d8\u6362\u6570\u636e\u96c6\u6765\u907f\u514d\u8fd9\u4e9b\u7ebf\u7d22\u8bcd\u51fa\u73b0\u65f6\uff0cBERT\u7684\u8868\u73b0\u6027\u80fd\u53d8\u5f97\u548c\u968f\u673a\u731c\u6d4b\u65e0\u5f02\u3002\u8fd9\u662f\u4e00\u4e2a\u660e\u663e\u7684\u4f9d\u9760\u6377\u5f84\u5b66\u4e60<\/span>\uff08shortcut learning\uff09<\/span>\u7684\u4f8b\u5b50\u2014\u2014\u4e00\u4e2a\u5728\u673a\u5668\u5b66\u4e60\u4e2d\u7ecf\u5e38\u88ab\u63d0\u53ca\u7684\u73b0\u8c61\uff0c\u5373\u5b66\u4e60\u7cfb\u7edf\u901a\u8fc7\u5206\u6790\u6570\u636e\u96c6\u4e2d\u7684\u4f2a\u76f8\u5173\u6027\uff0c\u800c\u4e0d\u662f\u901a\u8fc7\u7c7b\u4eba\u7406\u89e3<\/strong><\/span>\uff08humanlike understanding\uff09<\/span>\uff0c\u6765\u83b7\u5f97\u5728\u7279\u5b9a\u57fa\u51c6\u4efb\u52a1\u4e0a\u7684\u826f\u597d\u8868\u73b0<\/strong>[32-35]\u3002<\/span><\/p>\n

\u901a\u5e38\u60c5\u51b5\u4e0b\uff0c\u8fd9\u79cd\u76f8\u5173\u6027\u5bf9\u4e8e\u6267\u884c\u76f8\u540c\u4efb\u52a1\u7684\u4eba\u7c7b\u6765\u8bf4\u8868\u73b0\u5f97\u5e76\u4e0d\u660e\u663e\u3002\u867d\u7136\u6377\u5f84\u5b66\u4e60\u73b0\u8c61\u5728\u8bc4\u4f30\u8bed\u8a00\u7406\u89e3\u548c\u5176\u4ed6\u4eba\u5de5\u667a\u80fd\u6a21\u578b\u7684\u4efb\u52a1\u4e2d\u5df2\u7ecf\u88ab\u53d1\u73b0\uff0c\u4f46\u4ecd\u53ef\u80fd\u5b58\u5728\u5f88\u591a\u672a\u88ab\u53d1\u73b0\u7684\u201c\u6377\u5f84\u201d\u5b58\u5728\u3002\u50cf\u8c37\u6b4c\u7684LaMDA\u548cPaLM\u8fd9\u79cd\u62e5\u6709\u5343\u4ebf\u53c2\u6570\u89c4\u6a21\u3001\u5728\u8fd1\u4e07\u4ebf\u7684\u6587\u672c\u6570\u636e\u4e0a\u8fdb\u884c\u8bad\u7ec3\u7684\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\uff0c\u62e5\u6709\u5f3a\u5927\u7684\u7f16\u7801\u6570\u636e\u76f8\u5173\u6027\u7684\u80fd\u529b\u3002\u56e0\u6b64\uff0c\u7528\u4e8e\u8bc4\u4f30\u4eba\u7c7b\u7406\u89e3\u80fd\u529b\u7684\u57fa\u51c6\u4efb\u52a1\u6216\u8bb8\u5bf9\u8fd9\u7c7b\u6a21\u578b\u8bc4\u4f30\u6765\u8bf4\u5e76\u4e0d\u9002\u7528[36-38]\u3002\u5bf9\u4e8e\u5927\u89c4\u6a21LLMs<\/span>\uff08\u4ee5\u53caLLMs\u53ef\u80fd\u7684\u884d\u751f\u6a21\u578b\uff09<\/span>\u6765\u8bf4\uff0c\u901a\u8fc7\u590d\u6742\u7684\u7edf\u8ba1\u76f8\u5173\u6027\u8ba1\u7b97\u80fd\u591f\u8ba9\u6a21\u578b\u7ed5\u5f00\u7c7b\u4eba\u7406\u89e3\u80fd\u529b\uff0c\u83b7\u5f97\u8fd1\u4e4e\u5b8c\u7f8e\u7684\u6a21\u578b\u8868\u73b0<\/span>\u3002<\/span>
<\/p>\n

\u867d\u7136\u201c\u7c7b\u4eba\u7406\u89e3\u201d\u4e00\u8bcd\u6ca1\u6709\u4e25\u683c\u7684\u5b9a\u4e49\uff0c\u4f46\u5b83\u672c\u8d28\u4e0a\u5e76\u4e0d\u662f\u57fa\u4e8e\u5f53\u4e0bLLMs\u6240\u5b66\u4e60\u7684\u8fd9\u7c7b\u5e9e\u5927\u7684\u7edf\u8ba1\u6a21\u578b\uff1b\u76f8\u53cd\uff0c\u5b83\u57fa\u4e8e\u6982\u5ff5\u2014\u2014\u5916\u90e8\u7c7b\u522b\u3001\u60c5\u51b5\u548c\u4e8b\u4ef6\u7684\u5185\u90e8\u5fc3\u667a\u6a21\u578b\uff0c\u4ee5\u53ca\u4eba\u7c7b\u81ea\u8eab\u7684\u5185\u90e8\u72b6\u6001\u548c\u201c\u81ea\u6211\u201d\u7684\u5185\u90e8\u5fc3\u667a\u6a21\u578b\u3002\u5bf9\u4e8e\u4eba\u7c7b\u6765\u8bf4\uff0c\u7406\u89e3\u8bed\u8a00<\/span>\uff08\u4ee5\u53ca\u5176\u4ed6\u975e\u8bed\u8a00\u4fe1\u606f\uff09<\/span>\u4f9d\u8d56\u4e8e\u5bf9\u8bed\u8a00<\/span>\uff08\u6216\u5176\u4ed6\u4fe1\u606f\uff09<\/span>\u8868\u8fbe\u4e4b\u5916\u7684\u6982\u5ff5\u7684\u638c\u63e1\uff0c\u5e76\u975e\u5c40\u9650\u4e8e\u7406\u89e3\u8bed\u8a00\u7b26\u53f7\u7684\u7edf\u8ba1\u5c5e\u6027\u3002\u4e8b\u5b9e\u4e0a\uff0c\u5728\u8ba4\u77e5\u79d1\u5b66\u9886\u57df\u7684\u8fc7\u5f80\u7814\u7a76\u5386\u53f2\u4e2d\uff0c\u4e00\u76f4\u5f3a\u8c03\u5bf9\u6982\u5ff5\u672c\u8d28\u7684\u7406\u89e3\u4ee5\u53ca\u7406\u89e3\u529b\u662f\u5982\u4f55\u4ece\u6761\u7406\u6e05\u6670\u3001\u5c42\u6b21\u5206\u660e\u4e14\u5305\u542b\u6f5c\u5728\u56e0\u679c\u5173\u7cfb\u7684\u6982\u5ff5\u4e2d\u4ea7\u751f\u7684<\/span>\u3002\u8fd9\u79cd\u7406\u89e3\u529b\u6a21\u578b\u5e2e\u52a9\u4eba\u7c7b\u5bf9\u8fc7\u5f80\u77e5\u8bc6\u548c\u7ecf\u9a8c\u8fdb\u884c\u62bd\u8c61\u5316\u4ee5\u505a\u51fa\u7a33\u5065\u7684\u9884\u6d4b\u3001\u6982\u62ec\u548c\u7c7b\u6bd4\uff1b\u6216\u662f\u8fdb\u884c\u7ec4\u5408\u63a8\u7406\u3001\u53cd\u4e8b\u5b9e\u63a8\u7406\uff1b\u6216\u662f\u79ef\u6781\u5e72\u9884\u73b0\u5b9e\u4e16\u754c\u4ee5\u68c0\u9a8c\u5047\u8bbe\uff1b\u53c8\u6216\u662f\u5411\u4ed6\u4eba\u9610\u8ff0\u81ea\u5df1\u6240\u7406\u89e3\u7684\u5185\u5bb9\u3002<\/span><\/p>\n

\u6beb\u65e0\u7591\u95ee\uff0c\u5c3d\u7ba1\u6709\u4e9b\u89c4\u6a21\u8d8a\u6765\u8d8a\u5927\u7684LLMs\u96f6\u661f\u5730\u8868\u73b0\u51fa\u8fd1\u4f3c\u4eba\u7c7b\u7684\u7406\u89e3\u80fd\u529b\uff0c\u4f46\u5f53\u524d\u7684\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u5e76\u4e0d\u5177\u5907\u8fd9\u4e9b\u80fd\u529b\uff0c\u5305\u62ec\u6700\u524d\u6cbf\u7684LLMs\u3002\u6709\u4eba\u8ba4\u4e3a\uff0c\u8fd9\u79cd\u7406\u89e3\u80fd\u529b\u80fd\u591f\u8d4b\u4e88\u4eba\u7c7b\u7eaf\u7edf\u8ba1\u6a21\u578b\u65e0\u6cd5\u83b7\u5f97\u7684\u80fd\u529b\u3002\u5c3d\u7ba1\u5927\u6a21\u578b\u5c55\u73b0\u51fa\u4e86\u975e\u51e1\u7684\u5f62\u5f0f\u8bed\u8a00\u80fd\u529b<\/span><\/span>\uff08formal linguistic competence\uff09<\/span>\u2014\u2014\u5373\u4ea7\u751f\u8bed\u6cd5\u6d41\u5229\u3001\u7c7b\u4eba\u8bed\u8a00\u7684\u80fd\u529b\uff0c\u5b83\u4ecd\u7136\u7f3a\u4e4f\u57fa\u4e8e\u6982\u5ff5\u7406\u89e3\u7684\u7c7b\u4eba\u529f\u80fd\u8bed\u8a00\u80fd\u529b<\/span><\/span>\uff08humanlike functional language abilities\uff09<\/span>\u2014\u2014\u5373\u5728\u73b0\u5b9e\u4e16\u754c\u4e2d\u6b63\u786e\u7406\u89e3\u548c\u4f7f\u7528\u8bed\u8a00\u7684\u80fd\u529b\u3002\u6709\u8da3\u7684\u662f\uff0c\u7269\u7406\u5b66\u7814\u7a76\u4e2d\u4e5f\u6709\u7c7b\u4f3c\u7684\u73b0\u8c61\uff0c\u5373\u6570\u5b66\u6280\u6cd5\u7684\u6210\u529f\u8fd0\u7528\u548c\u8fd9\u79cd\u529f\u80fd\u7406\u89e3\u80fd\u529b\u4e4b\u95f4\u7684\u77db\u76fe<\/span>\u3002\u4f8b\u5982\uff0c\u4e00\u76f4\u4ee5\u6765\u5173\u4e8e\u91cf\u5b50\u529b\u5b66\u7684\u4e00\u4e2a\u4e89\u8bae\u662f\uff0c\u5b83\u63d0\u4f9b\u4e86\u4e00\u79cd\u6709\u6548\u7684\u8ba1\u7b97\u65b9\u6cd5\uff0c\u800c\u6ca1\u6709\u63d0\u4f9b\u6982\u5ff5\u6027\u7406\u89e3\u3002<\/span><\/p>\n

\u5173\u4e8e\u6982\u5ff5\u7684\u672c\u8d28\u7406\u89e3\u4e00\u76f4\u4ee5\u6765\u662f\u5b66\u754c\u4e89\u8bba\u7684\u4e3b\u9898\u4e4b\u4e00\u3002\u5bf9\u4e8e\u6982\u5ff5\u5728\u591a\u5927\u7a0b\u5ea6\u4e0a\u662f\u9886\u57df\u7279\u5b9a\u7684\u548c\u5148\u5929\u7684\uff0c\u800c\u4e0d\u662f\u66f4\u901a\u7528\u7684\u548c\u4e60\u5f97\u7684[55-60]\uff0c\u6216\u8005\u6982\u5ff5\u5728\u591a\u5927\u7a0b\u5ea6\u4e0a\u662f\u57fa\u4e8e\u5177\u8c61\u9690\u55bb\u7684\uff0c\u5e76\u901a\u8fc7\u52a8\u6001\u7684\u3001\u57fa\u4e8e\u60c5\u5883\u7684\u6a21\u62df\u5728\u5927\u8111\u4e2d\u5448\u73b0[64]\uff0c\u53c8\u6216\u8005\u6982\u5ff5\u5728\u4f55\u79cd\u6761\u4ef6\u4e0b\u662f\u7531\u8bed\u8a00[65\u201367]\u3001\u793e\u4f1a\u5b66\u4e60[68\u201370]\u548c\u6587\u5316\u652f\u6491\u7684[71\u201373]\uff0c\u7814\u7a76\u4eba\u5458\u5728\u8fd9\u4e9b\u65b9\u9762\u5b58\u5728\u5206\u6b67\u3002<\/span><\/p>\n

\u5c3d\u7ba1\u5b58\u5728\u4ee5\u4e0a\u4e89\u8bba\uff0c\u6982\u5ff5<\/strong>\u2014\u2014\u5c31\u50cf\u524d\u6587\u6240\u8ff0\u7684\u90a3\u6837\u4ee5\u56e0\u679c\u5fc3\u667a\u6a21\u578b\u7684\u5f62\u5f0f\u5b58\u5728\u2014\u2014\u4e00\u76f4\u4ee5\u6765\u88ab\u8ba4\u4e3a\u662f\u4eba\u7c7b\u8ba4\u77e5\u80fd\u529b\u7684\u7406\u89e3\u5355\u5143<\/strong>\u3002\u6beb\u65e0\u7591\u95ee\uff0c\u7eb5\u89c2\u4eba\u7c7b\u7406\u89e3\u80fd\u529b\u7684\u53d1\u5c55\u8f68\u8ff9\uff0c\u4e0d\u8bba\u662f\u4e2a\u4eba\u7406\u89e3\u8fd8\u662f\u96c6\u4f53\u7406\u89e3\uff0c\u90fd\u53ef\u4ee5\u62bd\u8c61\u4e3a\u5bf9\u4e16\u754c\u8fdb\u884c\u9ad8\u5ea6\u538b\u7f29\u7684\u3001\u57fa\u4e8e\u56e0\u679c\u5173\u7cfb\u7684\u6a21\u578b\uff0c\u7c7b\u4f3c\u4e8e\u4ece\u6258\u52d2\u5bc6\u7684\u884c\u661f\u516c\u8f6c\u7406\u8bba\u5230\u5f00\u666e\u52d2\u7684\u692d\u5706\u8f68\u9053\u7406\u8bba\uff0c\u518d\u5230\u725b\u987f\u6839\u636e\u5f15\u529b\u5bf9\u884c\u661f\u8fd0\u52a8\u7684\u7b80\u660e\u548c\u56e0\u679c\u5173\u7cfb\u7684\u89e3\u91ca\u3002\u4e0e\u673a\u5668\u4e0d\u540c\u7684\u662f\uff0c\u4eba\u7c7b\u4f3c\u4e4e\u5728\u79d1\u5b66\u7814\u7a76\u4ee5\u53ca\u65e5\u5e38\u751f\u6d3b\u4e2d\u90fd\u6709\u8ffd\u6c42\u8fd9\u79cd\u7406\u89e3\u5f62\u5f0f\u7684\u5f3a\u70c8\u5185\u9a71\u529b\u3002\u6211\u4eec\u53ef\u4ee5\u5c06\u8fd9\u79cd\u52a8\u529b\u63cf\u8ff0\u4e3a\u9700\u8981\u5f88\u5c11\u7684\u6570\u636e\uff0c\u6781\u7b80\u7684\u6a21\u578b\uff0c\u660e\u786e\u7684\u56e0\u679c\u4f9d\u8d56\u6027\u548c\u5f3a\u5927\u7684\u673a\u68b0\u76f4\u89c9\u3002<\/span><\/p>\n

\u5173\u4e8eLLMs\u7406\u89e3\u80fd\u529b\u7684\u4e89\u8bba\u4e3b\u8981\u96c6\u4e2d\u4ee5\u4e0b\u51e0\u4e2a\u65b9\u9762\uff1a<\/span><\/p>\n

1\uff09\u8fd9\u4e9b\u6a21\u578b\u7cfb\u7edf\u7684\u7406\u89e3\u80fd\u529b\u662f\u5426\u4ec5\u4ec5\u4e3a\u4e00\u79cd\u7c7b\u522b\u9519\u8bef\uff1f<\/span>\uff08\u5373\uff0c\u5c06\u8bed\u8a00\u7b26\u53f7\u4e4b\u95f4\u7684\u8054\u7cfb\u6df7\u6dc6\u4e3a\u7b26\u53f7\u4e0e\u7269\u7406\u3001\u793e\u4f1a\u6216\u5fc3\u667a\u4f53\u9a8c\u4e4b\u95f4\u7684\u8054\u7cfb\uff09<\/span>\u3002\u7b80\u800c\u8a00\u4e4b\uff0c\u8fd9\u4e9b\u6a21\u578b\u7cfb\u7edf\u6c38\u8fdc\u65e0\u6cd5\u83b7\u5f97\u7c7b\u4eba\u7684\u7406\u89e3\u80fd\u529b\u5417\uff1f<\/span><\/p>\n

\u6216\u8005\uff0c\u76f8\u53cd\u5730\uff0c2\uff09\u8fd9\u4e9b\u6a21\u578b\u7cfb\u7edf<\/span>\uff08\u6216\u8005\u5b83\u4eec\u8fd1\u671f\u7684\u884d\u751f\u6a21\u578b\uff09<\/span>\u771f\u7684\u4f1a\u5728\u7f3a\u4e4f\u73b0\u5b9e\u4e16\u754c\u7ecf\u9a8c\u7684\u60c5\u51b5\u4e0b\uff0c\u521b\u9020\u51fa\u5bf9\u4eba\u7c7b\u7406\u89e3\u6765\u8bf4\u81f3\u5173\u91cd\u8981\u7684\u5927\u91cf\u7684\u57fa\u4e8e\u6982\u5ff5\u7684\u5fc3\u667a\u6a21\u578b\u5417\uff1f\u5982\u679c\u662f\u7684\u8bdd\uff0c\u589e\u5927\u6a21\u578b\u89c4\u6a21\u662f\u5426\u4f1a\u521b\u9020\u51fa\u66f4\u597d\u7684\u6982\u5ff5\uff1f<\/span><\/p>\n

\u6216\u8005\uff0c3\uff09\u5982\u679c\u8fd9\u4e9b\u6a21\u578b\u7cfb\u7edf\u65e0\u6cd5\u521b\u9020\u8fd9\u6837\u7684\u6982\u5ff5\uff0c\u90a3\u4e48\u5b83\u4eec\u96be\u4ee5\u60f3\u8c61\u7684\u5e9e\u5927\u7684\u7edf\u8ba1\u76f8\u5173\u6027\u7cfb\u7edf\u662f\u5426\u80fd\u4ea7\u751f\u4e0e\u4eba\u7c7b\u7406\u89e3\u529f\u80fd\u76f8\u5f53\u7684\u80fd\u529b\u5462\uff1f\u53c8\u6216\u8005\uff0c\u8fd9\u662f\u5426\u610f\u5473\u7740\u4eba\u7c7b\u65e0\u6cd5\u8fbe\u5230\u7684\u65b0\u5f62\u5f0f\u7684\u9ad8\u9636\u903b\u8f91\u80fd\u529b\u6210\u4e3a\u53ef\u80fd\uff1f\u4ece\u8fd9\u4e00\u89d2\u5ea6\u4e0a\u770b\uff0c\u5c06\u8fd9\u79cd\u76f8\u5173\u6027\u79f0\u4e3a\u201c\u4f2a\u76f8\u5173\u6027\u201d\u6216\u8d28\u7591\u201c\u6377\u5f84\u5b66\u4e60\u201d\u73b0\u8c61\u662f\u5426\u4ecd\u7136\u5408\u9002\uff1f\u5c06\u6a21\u578b\u7cfb\u7edf\u7684\u884c\u4e3a\u89c6\u4e3a\u4e00\u7cfb\u5217\u65b0\u5174\u7684\u3001\u975e\u4eba\u7c7b\u7684\u7406\u89e3\u6d3b\u52a8\uff0c\u800c\u4e0d\u662f\u201c\u6ca1\u6709\u7406\u89e3\u80fd\u529b\u201d\uff0c\u662f\u5426\u884c\u5f97\u901a\uff1f\u8fd9\u4e9b\u95ee\u9898\u5df2\u4e0d\u518d\u5c40\u9650\u4e8e\u62bd\u8c61\u7684\u54f2\u5b66\u63a2\u8ba8\uff0c\u800c\u662f\u6d89\u53ca\u5230\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u5728\u4eba\u7c7b\u65e5\u5e38\u751f\u6d3b\u4e2d\u626e\u6f14\u7684\u8d8a\u6765\u8d8a\u91cd\u8981\u7684\u89d2\u8272\u6240\u5e26\u6765\u7684\u80fd\u529b\u3001\u7a33\u5065\u6027\u3001\u5b89\u5168\u6027\u548c\u4f26\u7406\u65b9\u9762\u7684\u975e\u5e38\u73b0\u5b9e\u7684\u62c5\u5fe7\u3002<\/span><\/p>\n

\u867d\u7136\u5404\u6d3e\u7814\u7a76\u8005\u5bf9\u4e8e\u201cLLMs\u7406\u89e3\u80fd\u529b\u201d\u7684\u4e89\u8bba\u90fd\u6709\u81ea\u8eab\u7684\u89c1\u89e3\uff0c\u4f46\u76ee\u524d\u7528\u4e8e\u83b7\u5f97\u7406\u89e3\u6d1e\u5bdf\u529b\u7684\u57fa\u4e8e\u8ba4\u77e5\u79d1\u5b66\u7684\u65b9\u6cd5\u4e0d\u8db3\u4ee5\u56de\u7b54\u5173\u4e8eLLMs\u7684\u8fd9\u7c7b\u95ee\u9898\u3002\u4e8b\u5b9e\u4e0a\uff0c\u4e00\u4e9b\u7814\u7a76\u4eba\u5458\u5df2\u7ecf\u5c06\u5fc3\u7406\u6d4b\u8bd5\u5e94\u7528\u4e8eLLMs\uff0c\u8fd9\u4e9b\u6d4b\u8bd5\u6700\u521d\u662f\u7528\u6765\u8bc4\u4f30\u4eba\u7c7b\u7406\u89e3\u548c\u63a8\u7406\u673a\u5236\u7684\u3002\u53d1\u73b0LLMs\u5728\u67d0\u4e9b\u60c5\u51b5\u4e0b\u786e\u5b9e\u5728\u5fc3\u7406\u7406\u8bba\u6d4b\u8bd5[14, 75]\u4e2d\u8868\u73b0\u51fa\u7c7b\u4f3c\u4eba\u7c7b\u7684\u53cd\u5e94\uff0c\u4ee5\u53ca\u5728\u63a8\u7406\u8bc4\u4f30\u4e2d\u8868\u73b0\u51fa\u7c7b\u4f3c\u4eba\u7c7b\u7684\u80fd\u529b\u548c\u504f\u597d [76\u201378]\u3002\u867d\u7136\u8fd9\u79cd\u6d4b\u8bd5\u88ab\u8ba4\u4e3a\u662f\u8bc4\u4f30\u4eba\u7c7b\u901a\u7528\u80fd\u529b\u7684\u66ff\u4ee3\u6027\u6d4b\u8bd5\uff0c\u4f46\u5bf9\u4eba\u5de5\u667a\u80fd\u6a21\u578b\u7cfb\u7edf\u6765\u8bf4\u53ef\u80fd\u5e76\u975e\u5982\u6b64\u3002<\/span><\/p>\n


<\/span><\/p>\n

\n
\n
\n
\n

\n
\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

\u4e00\u79cd\u65b0\u5174\u7684\u7406\u89e3\u80fd\u529b<\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n

\u6b63\u5982\u524d\u6587\u6240\u63d0\u5230\u7684\uff0cLLMs\u6709\u4e00\u79cd\u96be\u4ee5\u89e3\u91ca\u7684\u80fd\u529b\uff0c\u53ef\u4ee5\u5728\u8bad\u7ec3\u6570\u636e\u548c\u8f93\u5165\u4e2d\u5b66\u4e60\u4fe1\u606f\u7b26\u53f7\u4e4b\u95f4\u7684\u76f8\u5173\u6027\uff0c\u5e76\u4e14\u53ef\u4ee5\u4f7f\u7528\u8fd9\u79cd\u76f8\u5173\u6027\u6765\u89e3\u51b3\u95ee\u9898\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u4eba\u7c7b\u4f3c\u4e4e\u5e94\u7528\u4e86\u53cd\u6620\u4ed6\u4eec\u73b0\u5b9e\u4e16\u754c\u7ecf\u9a8c\u7684\u88ab\u538b\u7f29\u7684\u6982\u5ff5\u3002\u5f53\u628a\u4e3a\u4eba\u7c7b\u8bbe\u8ba1\u7684\u5fc3\u7406\u6d4b\u8bd5\u5e94\u7528\u4e8eLLMs\u65f6\uff0c\u5176\u89e3\u91ca\u7ed3\u679c\u5f80\u5f80\u4f9d\u8d56\u4e8e\u5bf9\u4eba\u7c7b\u8ba4\u77e5\u7684\u5047\u8bbe\uff0c\u800c\u8fd9\u4e9b\u5047\u8bbe\u5bf9\u4e8e\u6a21\u578b\u6765\u8bf4\u53ef\u80fd\u6839\u672c\u4e0d\u6b63\u786e\u3002\u4e3a\u4e86\u53d6\u5f97\u8fdb\u5c55\uff0c\u79d1\u5b66\u5bb6\u4eec\u9700\u8981\u8bbe\u8ba1\u65b0\u7684\u57fa\u51c6\u4efb\u52a1\u548c\u7814\u7a76\u65b9\u6cd5\uff0c\u4ee5\u6df1\u5165\u4e86\u89e3\u4e0d\u540c\u7c7b\u578b\u7684\u667a\u80fd\u548c\u7406\u89e3\u673a\u5236\uff0c\u5305\u62ec\u6211\u4eec\u5df2\u7ecf\u521b\u9020\u7684\u201c\u5f02\u7c7b\u7684\u3001\u7c7b\u4f3c\u601d\u7ef4\u5b9e\u4f53\u201d<\/span>\uff08exotic, mind-like entities\uff09<\/span>[79] \u7684\u65b0\u5f62\u5f0f\uff0c\u6216\u8bb8\u6211\u4eec\u6b63\u5728\u8e0f\u4e0a\u901a\u5f80\u6316\u6398\u201c\u7406\u89e3\u201d\u672c\u8d28\u7684\u6b63\u786e\u9053\u8def\u4e0a[80, 81]\u3002<\/span>
<\/p>\n

\u968f\u7740\u5173\u4e8eLLMs\u7406\u89e3\u80fd\u529b\u7684\u8ba8\u8bba\u58f0\u97f3\u8d8a\u6765\u8d8a\u591a\uff0c\u4ee5\u53ca\u66f4\u591a\u6709\u80fd\u529b\u7684\u6a21\u578b\u7cfb\u7edf\u7684\u51fa\u73b0\uff0c\u8fd9\u4e00\u5207\u4f3c\u4e4e\u90fd\u5728\u5f3a\u8c03\u672a\u6765\u6709\u5fc5\u8981\u52a0\u5f3a\u5bf9\u4e8e\u667a\u80fd\u79d1\u5b66\u7684\u7814\u7a76\uff0c\u4ee5\u4fbf\u5bf9\u4eba\u7c7b\u548c\u673a\u5668\u7684\u66f4\u5e7f\u6cdb\u7406\u89e3\u6982\u5ff5\u8fdb\u884c\u7406\u89e3\u3002\u6b63\u5982\u795e\u7ecf\u79d1\u5b66\u5bb6Terrence Sejnowski \u6240\u6307\u51fa\u7684\uff0c\u201c\u4e13\u5bb6\u4eec\u5bf9LLMs\u667a\u80fd\u7684\u5206\u6b67\u8868\u660e\uff0c\u6211\u4eec\u57fa\u4e8e\u81ea\u7136\u667a\u80fd\u7684\u4f20\u7edf\u89c2\u5ff5\u662f\u4e0d\u591f\u5145\u5206\u7684\u3002[9]\u201d\u5982\u679cLLMs\u548c\u5176\u4ed6\u6a21\u578b\u6210\u529f\u5730\u5229\u7528\u4e86\u5f3a\u5927\u7684\u7edf\u8ba1\u76f8\u5173\u6027\uff0c\u4e5f\u8bb8\u4e5f\u53ef\u4ee5\u88ab\u8ba4\u4e3a\u662f\u4e00\u79cd\u65b0\u5174\u7684\u201c\u7406\u89e3\u201d\u80fd\u529b\uff0c\u4e00\u79cd\u80fd\u591f\u5b9e\u73b0\u975e\u51e1\u7684\u3001\u8d85\u4eba\u7684\u9884\u6d4b\u80fd\u529b<\/span>\u3002\u6bd4\u5982DeepMind\u7684AlphaZero\u548cAlphaFold\u6a21\u578b\u7cfb\u7edf [82, 83]\uff0c\u5b83\u4eec\u4f3c\u4e4e\u5206\u522b\u4e3a\u56fd\u9645\u8c61\u68cb\u548c\u86cb\u767d\u8d28\u7ed3\u6784\u9884\u6d4b\u9886\u57df\u5e26\u6765\u4e86\u4e00\u79cd\u6765\u81ea\u201c\u5916\u661f\u201d\u7684\u76f4\u89c9\u5f62\u5f0f[84, 85]\u3002<\/span><\/p>\n

\u56e0\u6b64\u53ef\u4ee5\u8fd9\u6837\u8bf4\uff0c\u8fd1\u5e74\u6765\u5728\u4eba\u5de5\u667a\u80fd\u9886\u57df\u51fa\u73b0\u4e86\u5177\u6709\u65b0\u5174\u7406\u89e3\u6a21\u5f0f\u7684\u673a\u5668\uff0c\u8fd9\u6216\u8bb8\u662f\u4e00\u4e2a\u66f4\u5927\u7684\u76f8\u5173\u6982\u5ff5\u52a8\u7269\u56ed<\/span>\uff08zoo of related concepts\uff09<\/span>\u4e2d\u7684\u65b0\u7269\u79cd\u3002\u968f\u7740\u6211\u4eec\u5728\u8ffd\u6c42\u667a\u80fd\u672c\u8d28\u7684\u8fc7\u7a0b\u4e2d\u6240\u53d6\u5f97\u7684\u7814\u7a76\u8fdb\u5c55\uff0c\u8fd9\u4e9b\u65b0\u5174\u7684\u7406\u89e3\u6a21\u5f0f\u5c06\u4e0d\u65ad\u6d8c\u73b0\u3002\u6b63\u5982\u4e0d\u540c\u7684\u7269\u79cd\u9002\u5e94\u4e8e\u4e0d\u540c\u7684\u73af\u5883\u4e00\u6837\uff0c\u6211\u4eec\u7684\u667a\u80fd\u7cfb\u7edf\u4e5f\u5c06\u66f4\u597d\u5730\u9002\u5e94\u4e8e\u4e0d\u540c\u7684\u95ee\u9898\u3002\u4f9d\u8d56\u5927\u91cf\u7684\u5386\u53f2\u7684\u7f16\u7801\u77e5\u8bc6<\/span>\uff08encoded knowledge\uff09<\/span>\u7684\u95ee\u9898<\/span>\uff08\u5f3a\u8c03\u6a21\u578b\u6027\u80fd\u8868\u73b0\uff09<\/span>\u5c06\u7ee7\u7eed\u9752\u7750\u5927\u89c4\u6a21\u7684\u7edf\u8ba1\u6a21\u578b\uff0c\u5982LLMs\uff0c\u800c\u90a3\u4e9b\u4f9d\u8d56\u6709\u9650\u77e5\u8bc6\u548c\u5f3a\u5927\u56e0\u679c\u673a\u5236\u7684\u95ee\u9898\u5c06\u66f4\u9752\u7750\u4eba\u7c7b\u667a\u80fd\u3002\u672a\u6765\u7684\u6311\u6218\u662f\u5f00\u53d1\u51fa\u65b0\u7684\u7814\u7a76\u65b9\u6cd5\uff0c\u4ee5\u8be6\u7ec6\u63ed\u793a\u4e0d\u540c\u667a\u80fd\u5f62\u5f0f\u7684\u7406\u89e3\u673a\u5236\uff0c\u8fa8\u522b\u5b83\u4eec\u7684\u4f18\u52bf\u548c\u5c40\u9650\u6027\uff0c\u5e76\u5b66\u4e60\u5982\u4f55\u6574\u5408\u8fd9\u4e9b\u4e0d\u540c\u7684\u8ba4\u77e5\u6a21\u5f0f\u3002<\/span><\/section>\n

<\/span><\/section>\n


<\/span><\/p>\n

\n
\n
\n

\u53c2\u8003\u6587\u732e<\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n

\n
\n
\n
\n

<\/span><\/p>\n

[1]M. Mitchell, Arti\ufb01cial intelligence hits the barrier of meaning. Information 10, 51 (2019).<\/span><\/p>\n

[2]R. Bommasani et al., On the opportunities and risks of foundation models. arXiv [Preprint] (2021). http:\/\/arxiv.org\/abs\/2108.07258(Accessed 7 March 2023).<\/span><\/p>\n

[3]B. Min et al., Recent advances in natural language processing via large pre-trained language models: A survey. arXiv [Preprint] (2021). http:\/\/arxiv.org\/abs\/2111.01243(Accessed 7 March 2023).<\/span><\/p>\n

[4]L. Ouyang et al., Training language models to follow instructions with human feedback. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2203.02155(Accessed 7 March 2023).<\/span><\/p>\n

[5]T. Brown et al., Language models are few-shot learners. Adv. Neural Inf. Process. Syst. 33, 1877-1901 (2020).<\/span><\/p>\n

[6]J.Schulman et al., ChatGPT: Optimizing language models for dialogue. UpToDate (2022). https:\/\/openai.com\/blog\/chatgpt. Accessed 7 March 2023.<\/span><\/p>\n

[7]A. Chowdhery et al., PaLM: Scaling language modeling with Pathways. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2204.02311(Accessed 7 March 2023).<\/span><\/p>\n

[8]J. Wei et al., Chain of thought prompting elicits reasoning in large language models (2022). http:\/\/arxiv.org\/abs\/2201.11903(Accessed 7 March 2023).<\/span><\/p>\n

[9]T. Sejnowski, Large language models and the reverse Turing test. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2207.14382(Accessed 7 March 2023).<\/span><\/p>\n

[10]J. Wei et al., Emergent abilities of large language models. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2206.07682(Accessed 7 March 2023).<\/span><\/p>\n

[11]N. de Freitas, 14 May 2022. https:\/\/twitter.com\/NandoDF\/status\/1525397036325019649. Accessed 7 March 2023.<\/span><\/p>\n

[12]A. Dimakis, 16 May 2022. https:\/\/twitter.com\/AlexGDimakis\/status\/1526388274348150784. Accessed 7 March 2023.<\/span><\/p>\n

[13]R. Thoppilan et al., LaMDA: Language models for dialog applications. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2201.08239(Accessed 7 March 2023).<\/span><\/p>\n

[14]B. A. y Arcas, Do large language models understand us? UpToDate (2021). http:\/\/tinyurl.com\/38t23n73. Accessed 7 March 2023.<\/span><\/p>\n

[15]B. A. y Arcas, Arti\ufb01cial neural networks are making strides towards consciousness. UpToDate (2022). http:\/\/tinyurl.com\/ymhk37uu. Accessed 7 March 2023.<\/span><\/p>\n

[16]C. D. Manning, Human language understanding and reasoning. Daedalus 151, 127\u2013 138 (2022).<\/span><\/p>\n

[17]S. T. Piantasodi, F. Hill, Meaning without reference in large language models. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2208.02957(Accessed 7 March 2023).<\/span><\/p>\n

[18]B. A. y Arcas, Can machines learn how to behave? UpToDate (2022). http:\/\/tinyurl.com\/mr4cb3dw(Accessed 7 March 2023).<\/span><\/p>\n

[19]E. M. Bender, A. Koller, Climbing towards NLU: On meaning, form, and understanding in the age of data” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (2020), pp. 5185\u20135198.<\/span><\/p>\n

[20]E. M. Bender, T. Gebru, A. McMillan-Major, S. Shmitchell, On the dangers of stochastic parrots: Can language models be too big? in Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (2021), pp. 610\u2013623.<\/span><\/p>\n

[21]G. Marcus, Nonsense on stilts. Substack, 12 June 2022. https:\/\/garymarcus.substack.com\/p\/nonsense-on-stilts.<\/span><\/p>\n

[22]J. Browning, Y. LeCun, AI and the limits of language. UpToDate (2022)https:\/\/www.noemamag.com\/ai-and-the-limits-of-language. Accessed 7 March 2023.<\/span><\/p>\n

[23]A. Gopnik, What AI still doesn\u2019t know how to do. UpToDate (2022).https:\/\/www.wsj.com\/articles\/what-ai-still-doesnt-know-how-to-do-11657891316. Accessed 7 March 2023.<\/span><\/p>\n

[24]D. R. Hofstadter, Fluid Concepts and Creative Analogies: Computer Models of the Fundamental Mechanisms of Thought (Basic Books, Inc., New York, NY, 1995).<\/span><\/p>\n

[25]J. Weizenbaum, Computer Power and Human Reason: From Judgment to Calculation (WH Freeman & Co, 1976).<\/span><\/p>\n

[26]J. Michael et al., What do NLP researchers believe? Results of the NLP community metasurvey. arXiv [Preprint] (2022).http:\/\/arxiv.org\/abs\/2208.12852(Accessed 7 March 2023).<\/span><\/p>\n

[27]A. Wang et al.,”GLUE: A multi-task benchmark and analysis platform for natural language understanding”in Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP (Association for Computational Linguistics, 2018), pp. 353-355.<\/span><\/p>\n

[28]A. Wang et al., SuperGLUE: A stickier benchmark for general-purpose language understanding systems. Adv. Neural Inf. Process. Syst. 32, 3266\u20133280 (2019).<\/span><\/p>\n

[29]I. Habernal, H. Wachsmuth, I. Gurevych, B. Stein, “The argument reasoning comprehension task: Identfication and reconstruction of implicit warrants” in Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (2018), pp. 1930\u20131940.<\/span><\/p>\n

[30]J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, “BERT: Pre-training of deep bidirectional transformers for language understanding” in Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (2019), pp. 4171\u20134186.<\/span><\/p>\n

[31]T. Niven, H.-Y. Kao, Probing neural network comprehension of natural language arguments” in Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (2019), pp. 4658\u20134664.<\/span><\/p>\n

[32]R. Geirhos et al., Shortcut learning in deep neural networks. Nat. Mach. Intell. 2, 665\u2013673 (2020).<\/span><\/p>\n

[33]S. Gururangan et al., “Annotation artifacts in natural language inference data” in Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (2018), pp. 107\u2013112.<\/span><\/p>\n

[34]S. Lapuschkin et al., Unmasking Clever Hans predictors and assessing what machines really learn. Nat. Commun. 10, 1\u20138 (2019).<\/span><\/p>\n

[35]R T. McCoy, E. Pavlick, T. Linzen, “Right for the wrong reasons: Diagnosing syntactic heuristics in natural language inference” in Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (2019), pp. 3428\u20133448.<\/span><\/p>\n

[36]S. R. Choudhury, A. Rogers, I. Augenstein, Machine reading, fast and slow: When do models ‘understand\u2019language? arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2209.07430(Accessed 7 March 2023).<\/span><\/p>\n

[37]M. Gardner et al., “Competency problems: On finding and removing artifacts in language data” in Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (2021).<\/span><\/p>\n

[38]T. Linzen, How can we accelerate progress towards human-like linguistic generalization? in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (2020), pp. 5210\u20135217.<\/span><\/p>\n

[39]C. Baumberger, C. Beisbart, G. Brun, “What is understanding? An overview of recent debates in epistemology and philosophy of science” in Explaining Understanding: New Perspectives from Epistemology and Philosophy ofScience (Routledge, 2017), pp. 1\u201334.<\/span><\/p>\n

[40]J. L. Kvanvig, “Knowledge, understanding, and reasons for belief” in The Oxford Handbook of Reasons and Normativity (Oxford University Press, 2018), pp. 685\u2013705.<\/span><\/p>\n

[41]M. B. Goldwater, D. Gentner, On the acquisition of abstract knowledge: Structural alignment and explication in learning causal system categories. Cognition 137, 137\u2013153 (2015).<\/span><\/p>\n

[42]A. Gopnik, \u201cCausal models and cognitive development\u201din Probabilistic and Causal Inference: The Works ofJudea Pearl, H. Geffner, R. Dechter, J. Y. Halpern, Eds. (Association for Computing Machinery, 2022), pp. 593\u2013604.<\/span><\/p>\n

[43]D. R. Hofstadter, E. Sander, Surfaces and Essences: Analogy as the Fuel and Fire of Thinking. Basic Books (2013).<\/span><\/p>\n

[44]F. C. Keil, Explanation and understanding. Ann. Rev. Psychol. 57, 227 (2006).<\/span><\/p>\n

[45]M. Lake, T. D. Ullman, J. B. Tenenbaum, S. J. Gershman, Building machines that learn and think like people. Behav. Brain Sci. 40 (2017).<\/span><\/p>\n

[46]S. A. Sloman, D. Lagnado, Causality in thought. Ann. Rev. Psychol. 66, 223\u2013247 (2015).<\/span><\/p>\n

[47]P. Smolensky, R. McCoy, R. Fernandez, M. Goldrick, J. Gao, Neurocompositional computing: From the central paradox of cognition to a new generation of AI systems. AI Mag. 43, 308\u2013322 (2022).<\/span><\/p>\n

[48]H. W. De Regt, Discussion note: Making sense of understanding. Philos. Sci. 71, 98\u2013109 (2004).<\/span><\/p>\n

[49]D. George, M. L\u00e1zaro-Gredilla, J. S. Guntupalli, From CAPTCHA to commonsense: How brain can teach us about arti\ufb01cial intelligence. Front. Comput. Neurosci. 14, 554097 (2020).<\/span><\/p>\n

[50]B. M. Lake, G. L. Murphy, Word meaning in minds and machines. Psychol. Rev. (2021).<\/span><\/p>\n

[51]J. Pearl, Theoretical impediments to machine learning with seven sparks from the causal revolution. arXiv [Preprint] (2018).http:\/\/arxiv.org\/abs\/1801.04016(Accessed 7 March 2023).<\/span><\/p>\n

[52]M. Strevens, No understanding without explanation. Stud. Hist. Philos. Sci. A. 44, 510\u2013515 (2013).<\/span><\/p>\n

[53]K. Mahowald et al., Dissociating language and thought in large language models: a cognitive perspective. arXiv [Preprint] (2023).http:\/\/arxiv.org\/abs\/2301.06627(Accessed 7 March 2023).<\/span><\/p>\n

[54]D. C. Krakauer, At the limits of thought. UpToDate (2020).https:\/\/aeon.co\/essays\/will-brains-or-algorithms-rule-the-kingdom-of-science. Accessed 7 March 2023.<\/span><\/p>\n

[55]S. Carey, \u201cOn the origin of causal understanding\u201din Causal Cognition: A Multidisciplinary Debate, D. Sperber, D. Premack, A. J. Premack, Eds. (Clarendon Press\/Oxford University Press, 1995), pp. 268\u2013308.<\/span><\/p>\n

[56]N. D. Goodman, T. D. Ullman, J. B. Tenenbaum, Learning a theory of causality. Psychol. Rev. 118, 110 (2011).<\/span><\/p>\n

[57]A. Gopnik, A uni\ufb01ed account of abstract structure and conceptual change: Probabilistic models and early learning mechanisms. Behav. Brain Sci. 34, 129 (2011).<\/span><\/p>\n

[58]J. M. Mandler, How to build a baby: II. Conceptual primitives. Psychol. Rev. 99, 587 (1992).<\/span><\/p>\n

[59]E. S. Spelke, K. D. Kinzler, Core knowledge. Dev. Sci. 10, 89\u201396 (2007).<\/span><\/p>\n

[60]H. M. Wellman, S. A. Gelman, Cognitive development: Foundational theories of core domains. Ann. Rev. Psychol. 43, 337\u2013375 (1992).<\/span><\/p>\n

[61]R. W. Gibbs, Metaphor Wars (Cambridge University Press, 2017).<\/span><\/p>\n

[62]G. Lakoff, M. Johnson, The metaphorical structure of the human conceptual system. Cogn. Sci. 4, 195\u2013208 (1980).<\/span><\/p>\n

[63]G. L. Murphy, On metaphoric representation. Cognition 60, 173\u2013204 (1996).<\/span><\/p>\n

[64]L. W. Barsalou et al., Grounded cognition. Ann. Rev. Psychol. 59, 617\u2013645 (2008).<\/span><\/p>\n

[65]J. G. De Villiers, P. A. de Villiers, The role of language in theory of mind development. Topics Lang. Disorders 34, 313\u2013328 (2014).<\/span><\/p>\n

[66]G. Dove, More than a scaffold: Language is a neuroenhancement. Cogn. Neuropsychol. 37, 288\u2013311 (2020).<\/span><\/p>\n

[67]G. Lupyan, B. Bergen, How language programs the mind. Topics Cogn. Sci. 8, 408\u2013424 (2016).<\/span><\/p>\n

[68]N. Akhtar, M. Tomasello, \u201cThe social nature of words and word learning\u201din Becoming a Word Learner: A Debate on Lexical Acquisition (Oxford University Press, 2000), pp. 115\u2013 135.<\/span><\/p>\n

[69]S. R. Waxman, S. A. Gelman, Early word-learning entails reference, not merely associations. Trends Cogn. Sci. 13, 258\u2013263 (2009).<\/span><\/p>\n

[70]S. A. Gelman, Learning from others: Children\u2019s construction of concepts. Ann. Rev. Psychol. 60, 115\u2013 140 (2009).<\/span><\/p>\n

[71]A. Bender, S. Beller, D. L. Medin, \u201cCausal cognition and culture\u201din The Oxford Handbook of Causal Reasoning (Oxford University Press, 2017), pp. 717\u2013738.<\/span><\/p>\n

[72]M. W. Morris, T. Menon, D. R. Ames,\u201dCulturally conferred conceptions of agency: A key to social perception of persons, groups, and other actors\u201din Personality and Social Psychology Review (Psychology Press, 2003), pp. 169\u2013182.<\/span><\/p>\n

[73]A. Norenzayan, R. E. Nisbett, Culture and causal cognition. Curr. Direc. Psychol. Sci. 9, 132\u2013 135 (2000).<\/span><\/p>\n

[74]A. Gopnik, H. M. Wellman, \u201cThe theory theory\u201din Domain Speci\ufb01city in Cognition and Culture (1994), pp. 257\u2013293.<\/span><\/p>\n

[75]S. Trott, C. Jones, T. Chang, J. Michaelov, B. Bergen, Do large language models know what humans know? arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2209.01515(Accessed 7 March 2023).<\/span><\/p>\n

[76]M. Binz, E. Schulz, Using cognitive psychology to understand GPT-3. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2206.14576(Accessed 7 March 2023).<\/span><\/p>\n

[77]I. Dasgupta et al., Language models show human-like content effects on reasoning. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2207.07051(Accessed 7 March 2023).<\/span><\/p>\n

[78]A. Laverghetta, A. Nighojkar, J. Mirzakhalov, J. Licato, \u201cPredicting human psychometric properties using computational language models\u201din Annual Meeting of the Psychometric Society (Springer, 2022), pp. 151\u2013169.<\/span><\/p>\n

[79]M. Shanahan, Talking about large language models. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2212.03551(Accessed 7 March 2023).<\/span><\/p>\n

[80]B. Z. Li, M. Nye, J. Andreas, \u201cImplicit representations of meaning in neural language models\u201d in Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics (2021), pp. 1813\u20131827.<\/span><\/p>\n

[81]C. Olsson et al., In-context learning and induction heads. arXiv [Preprint] (2022). http:\/\/arxiv.org\/abs\/2209.11895(Accessed 7 March 2023).<\/span><\/p>\n

[82]J. Jumper et al., Highly accurate protein structure prediction with AlphaFold. Nature 596, 583\u2013589 (2021).<\/span><\/p>\n

[83]D. Silver et al., Mastering chess and shogi by self-play with a general reinforcement learning algorithm. arXiv [Preprint] (2017).http:\/\/arxiv.org\/abs\/1712.01815(Accessed 7 March 2023).<\/span><\/p>\n

[84]D. T. Jones, J. M. Thornton, The impact of AlphaFold2 one year on. Nat. Methods 19, 15\u201320 (2022).<\/span><\/p>\n

[85]M. Sadler, N. Regan, Game changer: AlphaZero\u2019s Groundbreaking Chess Strategies and the Promise ofAI. Alkmaar (New in Chess, 2019).<\/span><\/p>\n


<\/span><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\n
\n

\uff08\u53c2\u8003\u6587\u732e\u53ef\u4e0a\u4e0b\u6ed1\u52a8<\/strong>\u67e5\u770b\uff09<\/p>\n<\/section>\n<\/section>\n


<\/section>\n

<\/section>\n
\n
\n
\n
\n

\u201c\u540eChatGPT\u201d\u8bfb\u4e66\u4f1a\u542f\u52a8<\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\n


<\/p>\n

2022\u5e7411\u670830\u65e5\uff0c\u4e00\u4e2a\u73b0\u8c61\u7ea7\u5e94\u7528\u7a0b\u5e8f\u8bde\u751f\u4e8e\u4e92\u8054\u7f51\uff0c\u8fd9\u5c31\u662fOpenAI\u5f00\u53d1\u7684ChatGPT\u3002\u4ece\u95ee\u7b54\u5230\u5199\u7a0b\u5e8f\uff0c\u4ece\u63d0\u53d6\u6458\u8981\u5230\u8bba\u6587\u5199\u4f5c\uff0cChatGPT\u5c55\u73b0\u51fa\u4e86\u591a\u6837\u5316\u7684\u901a\u7528\u667a\u80fd\u3002\u4e8e\u662f\uff0c\u5fae\u8f6f\u3001\u8c37\u6b4c\u3001\u767e\u5ea6\u3001\u963f\u91cc\u3001\u8baf\u98de\uff0c\u4e92\u8054\u7f51\u5927\u4f6c\u4eec\u7eb7\u7eb7\u6469\u62f3\u64e6\u638c\u51c6\u5907\u5165\u573a\u2026\u2026\u4f46\u662f\uff0c\u8bf7\u5148\u51b7\u9759\u4e00\u4e0b\u2026\u2026 \u73b0\u5728 all in \u5927\u8bed\u8a00\u6a21\u578b\u662f\u5426\u771f\u7684\u5408\u9002\uff1f\u8981\u77e5\u9053\uff0cChatGPT\u7684\u80cc\u540e\u5176\u5b9e\u5c31\u662f\u6df1\u5ea6\u5b66\u4e60+\u5927\u6570\u636e+\u5927\u6a21\u578b\uff0c\u800c\u8fd9\u4e9b\u8981\u7d20\u65e9\u57285\u5e74\u524d\u7684AlphaGo\u65f6\u671f\u5c31\u5df2\u7ecf\u5f00\u59cb\u706b\u70ed\u4e86\u30025\u5e74\u524d\u6ca1\u6709\u6293\u4f4f\u673a\u9047\uff0c\u73b0\u5728\u53c8\u51ed\u4ec0\u4e48\u53ef\u4ee5\u642d\u4e0a\u5927\u8bed\u8a00\u6a21\u578b\u8fd9\u8d9f\u5217\u8f66\u5462\uff1f<\/span><\/section>\n

<\/section>\n
\u96c6\u667a\u4ff1\u4e50\u90e8\u7279\u522b\u7ec4\u7ec7\u201c\u540e ChatGPT\u201d\u8bfb\u4e66\u4f1a<\/a>\uff0c\u7531\u5317\u5e08\u5927\u6559\u6388\u3001\u96c6\u667a\u4ff1\u4e50\u90e8\u521b\u59cb\u4eba\u5f20\u6c5f\u8001\u5e08\u8054\u5408\u8096\u8fbe\u3001\u674e\u5ae3\u7136\u3001\u5d14\u9e4f\u3001\u4faf\u6708\u6e90\u3001\u949f\u7ff0\u5ef7\u3001\u5362\u71da\u7b49\u591a\u4f4d\u8001\u5e08\u5171\u540c\u53d1\u8d77\uff0c\u65e8\u5728\u7cfb\u7edf\u6027\u5730\u68b3\u7406ChatGPT\u6280\u672f\uff0c\u5e76\u53d1\u73b0\u5176\u5f31\u70b9\u4e0e\u77ed\u677f\u3002\u672c\u7cfb\u5217\u8bfb\u4e66\u4f1a\u7ebf\u4e0a\u8fdb\u884c\uff0c2023\u5e743\u67083\u65e5\u5f00\u59cb\uff0c\u6bcf\u5468\u4e94\u665a\uff0c\u6b22\u8fce\u62a5\u540d\u4ea4\u6d41\u3002<\/span><\/section>\n


<\/p>\n<\/section>\n

<\/span><\/a><\/p>\n

\u8be6\u60c5\u8bf7\u89c1\uff1a<\/span><\/section>\n
\u201c\u540e ChatGPT\u201d\u8bfb\u4e66\u4f1a\u542f\u52a8\uff1a\u4ece\u901a\u7528\u4eba\u5de5\u667a\u80fd\u5230\u610f\u8bc6\u673a\u5668<\/a><\/section>\n

<\/section>\n

<\/span><\/section>\n
\n

\u63a8\u8350\u9605\u8bfb<\/span><\/strong><\/strong><\/p>\n

1. <\/strong><\/strong><\/strong><\/strong><\/a>\u4eba\u7c7b\u8bed\u8a00\u80fd\u529b\u7684\u81ea\u7136\u6f14\u5316\uff1a\u4e54\u59c6\u65af\u57fa\u5bf9\u9635\u8fbe\u5c14\u6587|\u300a\u8fbe\u5c14\u6587\u7684\u5371\u9669\u601d\u60f3\u300b<\/a><\/strong><\/strong><\/strong><\/strong><\/span><\/section>\n
2.<\/strong> <\/strong>AI\u4f55\u4ee5\u6d8c\u73b0\uff1a\u590d\u6742\u9002\u5e94\u7cfb\u7edf\u89c6\u89d2\u7684ChatGPT\u548c\u5927\u8bed\u8a00\u6a21\u578b<\/a> <\/strong><\/span><\/strong><\/section>\n
3.<\/strong><\/strong> <\/strong>\u201c\u610f\u8bc6\u673a\u5668\u201d\u521d\u63a2\uff1a\u5982\u4f55\u8ba9\u5927\u8bed\u8a00\u6a21\u578b\u5177\u5907\u81ea\u6211\u610f\u8bc6\uff1f<\/strong><\/a><\/span><\/section>\n
4. <\/strong><\/strong><\/strong><\/strong>Science\u524d\u6cbf\uff1a\u5927\u8bed\u8a00\u6a21\u578b\u6d8c\u73b0\u6f14\u5316\u4fe1\u606f\uff0c\u52a0\u901f\u86cb\u767d\u8d28\u7ed3\u6784\u9884\u6d4b<\/a><\/strong><\/strong><\/span><\/section>\n
5. \u300a\u5f20\u6c5f\u00b7\u590d\u6742\u79d1\u5b66\u524d\u6cbf27\u8bb2\u300b\u5b8c\u6574\u4e0a\u7ebf\uff01<\/strong><\/a><\/strong><\/strong><\/strong><\/section>\n
6. \u6210\u4e3a\u96c6\u667aVIP\uff0c\u89e3\u9501\u5168\u7ad9\u8bfe\u7a0b\uff0f\u8bfb\u4e66\u4f1a<\/span><\/strong><\/a><\/strong><\/strong><\/strong><\/strong><\/section>\n
7. <\/strong><\/strong><\/strong>\u52a0\u5165\u96c6\u667a\uff0c\u4e00\u8d77\u590d\u6742\uff01<\/strong><\/a><\/strong><\/strong><\/strong><\/span><\/section>\n

<\/strong><\/strong><\/strong><\/span><\/section>\n


<\/p>\n


<\/p>\n<\/section>\n

\u70b9\u51fb\u201c\u9605\u8bfb\u539f\u6587\u201d\uff0c\u62a5\u540d\u8bfb\u4e66\u4f1a<\/span><\/strong><\/section>\n

<\/mp-style-type><\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

\u5bfc\u8bed \u5c3d\u7ba1\u5927\u8bed\u8a00\u6a21\u578b\u8868\u73b0\u51fa\u8fd1\u4f3c\u4eba\u7c7b\u7684\u7406\u89e3\u80fd\u529b\uff0c\u4f46 AI \u7cfb\u7edf\u771f\u7684\u53ef\u4ee5\u50cf\u4eba\u7c7b\u4e00\u6837\u7406\u89e3\u8bed\u8a00\u5417\uff1f\u673a\u5668\u7406\u89e3\u7684\u6a21\u5f0f\u5fc5\u987b\u548c\u4eba\u7c7b\u7406\u89e3\u76f8\u540c\u5417\uff1f\u8fd1\u65e5\uff0c\u5723\u5854\u83f2\u7814\u7a76\u6240\u524d\u6240\u957f\u514b\u62c9\u8003\u5c14\u548c\u7814\u7a76\u5458\u6885\u62c9\u5c3c\u00b7\u7c73\u6b47\u5c14\u5728 PNAS \u53d1\u8868\u6587\u7ae0\uff0c\u63a2\u8ba8\u5927\u578b\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u662f\u5426\u80fd\u591f\u4ee5\u7c7b\u4f3c\u4eba\u7c7b\u7684\u65b9\u5f0f\u7406\u89e3\u8bed\u8a00\u53ca\u5176\u6240\u7f16\u7801\u7684\u7269\u7406\u548c\u793e\u4f1a\u60c5\u5883\u3002 …<\/p>\n","protected":false},"author":0,"featured_media":41973,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/41975"}],"collection":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=41975"}],"version-history":[{"count":0,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/41975\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/41973"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=41975"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=41975"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=41975"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=41975"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}