{"id":48540,"date":"2024-02-26T18:54:16","date_gmt":"2024-02-26T10:54:16","guid":{"rendered":"https:\/\/swarma.org\/?p=48540"},"modified":"2024-02-26T18:54:16","modified_gmt":"2024-02-26T10:54:16","slug":"%e2%80%8b%e5%8c%97%e5%a4%a7%e5%8f%91%e8%a1%a8-ai-alignment-%e7%bb%bc%e8%bf%b0%ef%bc%9a%e7%a1%ae%e4%bf%9dai%e4%b8%8e%e4%ba%ba%e7%b1%bb%e4%bb%b7%e5%80%bc%e8%a7%82%e4%b8%80%e8%87%b4%e7%9a%84%e5%9b%9b","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=48540","title":{"rendered":"\u200b\u5317\u5927\u53d1\u8868 AI Alignment \u7efc\u8ff0\uff1a\u786e\u4fddAI\u4e0e\u4eba\u7c7b\u4ef7\u503c\u89c2\u4e00\u81f4\u7684\u56db\u4e2a\u5173\u952e\u8bbe\u8ba1\u539f\u5219"},"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
\u8fd1\u671f\uff0c\u6765\u81ea\u5317\u5927\u7684\u7814\u7a76\u56e2\u961f\u6574\u7406\u4e86\u4e00\u4efd\u8be6\u5c3d\u7684 AI \u5bf9\u9f50\u6700\u65b0\u7efc\u8ff0\uff0c\u672c\u6587\u6982\u8ff0\u4e86\u5f53\u524d\u4eba\u5de5\u667a\u80fd\u5bf9\u9f50\u7814\u7a76\u7684\u5168\u8c8c\uff0c\u4f9d\u636e\u56db\u4e2a\u5173\u952e\u539f\u5219\uff0c\u5c06\u5176\u5206\u89e3\u4e3a\u4e24\u4e2a\u5173\u952e\u7ec4\u6210\u90e8\u5206\uff1a\u524d\u5411\u5bf9\u9f50\u548c\u540e\u5411\u5bf9\u9f50\uff0c\u4ece\u800c\u8fdb\u884c\u66f4\u5168\u9762\u7684\u8ba8\u8bba\u3002\u5f53\u524d\u7684\u7814\u7a76\u548c\u5b9e\u8df5\u5c06\u8fd9\u4e9b\u76ee\u6807\u7eb3\u5165\u4e86\u53cd\u9988\u5b66\u4e60\u3001\u5206\u5e03\u504f\u79fb\u5b66\u4e60\u3001\u4fdd\u8bc1\u548c\u6cbb\u7406\u7b49\u56db\u4e2a\u9886\u57df\u3002AI \u5bf9\u9f50\u7684\u76ee\u6807\u4e0d\u4ec5\u4ec5\u662f\u4e3a\u4e86\u907f\u514d AI \u7cfb\u7edf\u7684\u4e0d\u826f\u884c\u4e3a\uff0c\u66f4\u91cd\u8981\u7684\u662f\u786e\u4fdd\u5176\u5728\u6267\u884c\u4efb\u52a1\u65f6\u7b26\u5408\u4eba\u7c7b\u7684\u610f\u56fe\u548c\u4ef7\u503c\u89c2\u3002<\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/section>\n

<\/section>\n
\u7814\u7a76\u9886\u57df\uff1aAI\u5bf9\u9f50\uff0cRICE\u539f\u5219\uff0c\u53cd\u9988\u5b66\u4e60\uff0c\u5206\u5e03\u504f\u79fb\u5b66\u4e60<\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/strong><\/section>\n<\/section>\n
\n
<\/span><\/strong><\/section>\n<\/section>\n
\n
<\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/strong><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/strong><\/span><\/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
<\/section>\n<\/section>\n
\n
 \u667a\u5546\u6389\u4e86\u4e00\u5730\u3001\u5356\u840c\u9171<\/span><\/span><\/span> | \u4f5c\u8005<\/span><\/strong><\/section>\n
\u5915\u5c0f\u7476\u79d1\u6280\u8bf4<\/span><\/span> <\/span>| \u6765\u6e90<\/span><\/strong><\/span><\/strong><\/section>\n

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

<\/section>\n
\n
\u8fd1\u671f\u8ba9\u4e92\u8054\u7f51\u6253\u5de5\u4eba\u5f88\u6709\u5171\u9e23\u7684\u8bcd\u83ab\u8fc7\u4e8e\u201c\u5bf9\u9f50\u9897\u7c92\u5ea6\u201d\u4e86\uff0c\u4f46\u201c\u5bf9\u9f50<\/span>\uff08Alignment\uff09<\/span>\u201d\u8fd9\u4e00\u6982\u5ff5\u96be\u9053\u53ea\u51fa\u73b0\u5728\u6253\u5de5\u4eba\u7684\u4ea4\u6d41\u573a\u666f\u4e2d\u5417\uff1fNo\uff01\u968f\u7740\u4eba\u5de5\u667a\u80fd<\/span>\uff08AI\uff09<\/span>\u6280\u672f\u7684\u5feb\u901f\u53d1\u5c55\uff0cAI \u7cfb\u7edf\u5728\u793e\u4f1a\u5404\u4e2a\u9886\u57df\u7684\u5e94\u7528\u65e5\u76ca\u5e7f\u6cdb\uff0c\u4eba\u7c7b\u548c AI \u4e4b\u95f4\u7684\u4ea4\u6d41\u4e5f\u9700\u8981\u5bf9\u9f50\u3002AI \u5bf9\u9f50\u7684\u76ee\u6807\u662f\u4f7f AI \u7cfb\u7edf\u7684\u884c\u4e3a\u4e0e\u4eba\u7c7b\u610f\u56fe\u548c\u4ef7\u503c\u4fdd\u6301\u4e00\u81f4\u3002<\/span><\/section>\n

<\/span><\/section>\n
\u4ece\u81ea\u52a8\u9a7e\u9a76\u5230\u533b\u7597\u8bca\u65ad\uff0c\u518d\u5230\u91d1\u878d\u5206\u6790\u4e0e\u5ba2\u6237\u670d\u52a1\uff0cAI \u7cfb\u7edf\u7684\u80fd\u529b\u5728\u4e0d\u65ad\u63d0\u5347\uff0c\u4f7f\u5f97\u5b83\u4eec\u80fd\u591f\u5904\u7406\u66f4\u52a0\u590d\u6742\u548c\u9ad8\u98ce\u9669\u7684\u4efb\u52a1\u3002\u4f46\u4e00\u4e9b AI \u7cfb\u7edf\u5c55\u73b0\u51fa\u7684\u4e0d\u826f\u884c\u4e3a\uff0c\u5f15\u53d1\u4e86\u5bf9 AI \u7cfb\u7edf\u6f5c\u5728\u5371\u5bb3\u7684\u62c5\u5fe7\uff0c\u4e00\u4e2a\u5173\u952e\u95ee\u9898\u4e5f\u968f\u4e4b\u800c\u6765\uff1a\u5982\u4f55\u786e\u4fdd AI \u7cfb\u7edf\u7684\u884c\u4e3a\u4e0e\u4eba\u7c7b\u7684\u610f\u56fe\u548c\u4ef7\u503c\u89c2\u4fdd\u6301\u4e00\u81f4\uff1f<\/span><\/section>\n

<\/span><\/section>\n
<\/p>\n
AI \u5b58\u5728\u7684\u5371\u9669\u80fd\u529b<\/span><\/section>\n<\/figure>\n

<\/section>\n
AI \u7cfb\u7edf\u4e0e\u4eba\u7c7b\u4ef7\u503c\u89c2\u4e00\u81f4\u6027\u7684\u91cd\u8981\u6027\u4e0d\u4ec5\u4f53\u73b0\u5728\u907f\u514d\u7cfb\u7edf\u505a\u51fa\u6709\u6096\u4e8e\u4eba\u7c7b\u610f\u613f\u7684\u884c\u4e3a\u4e0a\uff0c\u66f4\u5173\u4e4e\u4e8e\u793e\u4f1a\u7684\u957f\u671f\u7a33\u5b9a\u548c\u53d1\u5c55\u3002\u56e0\u6b64\uff0c\u7814\u7a76\u548c\u5f00\u53d1\u80fd\u591f\u7406\u89e3\u5e76\u5c0a\u91cd\u4eba\u7c7b\u4ef7\u503c\u89c2\u7684 AI \u7cfb\u7edf\uff0c\u5df2\u7ecf\u6210\u4e3a AI \u9886\u57df\u7684\u4e00\u4e2a\u91cd\u8981\u8bae\u9898\u3002<\/span><\/section>\n

<\/span><\/section>\n
\u8fd1\u671f\uff0c\u6765\u81ea\u5317\u5927\u7684\u7814\u7a76\u56e2\u961f\u6574\u7406\u4e86\u4e00\u4efd\u8be6\u5c3d\u7684 AI \u5bf9\u9f50\u6700\u65b0\u7efc\u8ff0\uff0c\u672c\u6587\u6982\u8ff0\u4e86\u5f53\u524d\u4eba\u5de5\u667a\u80fd\u5bf9\u9f50\u7814\u7a76\u7684\u5168\u8c8c\uff0c\u4f9d\u636e\u56db\u4e2a\u5173\u952e\u539f\u5219\uff0c\u5c06\u5176\u5206\u89e3\u4e3a\u4e24\u4e2a\u5173\u952e\u7ec4\u6210\u90e8\u5206\uff1a\u524d\u5411\u5bf9\u9f50\u548c\u540e\u5411\u5bf9\u9f50\uff0c\u4ece\u800c\u8fdb\u884c\u66f4\u5168\u9762\u7684\u8ba8\u8bba\u3002\u5f53\u524d\u7684\u7814\u7a76\u548c\u5b9e\u8df5\u5c06\u8fd9\u4e9b\u76ee\u6807\u7eb3\u5165\u4e86\u53cd\u9988\u5b66\u4e60\u3001\u5206\u5e03\u504f\u79fb\u5b66\u4e60\u3001\u4fdd\u8bc1\u548c\u6cbb\u7406\u7b49\u56db\u4e2a\u9886\u57df\u3002AI \u5bf9\u9f50\u7684\u76ee\u6807\u4e0d\u4ec5\u4ec5\u662f\u4e3a\u4e86\u907f\u514d AI \u7cfb\u7edf\u7684\u4e0d\u826f\u884c\u4e3a\uff0c\u66f4\u91cd\u8981\u7684\u662f\u786e\u4fdd\u5176\u5728\u6267\u884c\u4efb\u52a1\u65f6\u7b26\u5408\u4eba\u7c7b\u7684\u610f\u56fe\u548c\u4ef7\u503c\u89c2\u3002<\/span><\/section>\n

<\/span><\/section>\n
\n
\n
\n
\u8bba\u6587\u9898\u76ee:
AI Alignment: A Comprehensive Survey<\/span><\/section>\n
\u8bba\u6587\u94fe\u63a5:
https:\/\/arxiv.org\/abs\/2310.19852<\/span><\/section>\n
\u535a\u5ba2\u5730\u5740:
http:\/\/www.alignmentsurvey.com<\/span><\/section>\n
GitHub \u5730\u5740:
https:\/\/github.com\/PKU-Alignment<\/span><\/section>\n<\/section>\n<\/section>\n<\/blockquote>\n

<\/span><\/section>\n
\u8fd1\u5e74\u6765\u7814\u7a76\u4eba\u5458\u5bf9\u5927\u578b\u8bed\u8a00\u6a21\u578b<\/span>\uff08LLM\uff09<\/span>\u548c\u5f3a\u5316\u5b66\u4e60<\/span>\uff08RL\uff09<\/span>\u7684\u6df1\u5165\u63a2\u7d22\uff0c\u91cd\u65b0\u70b9\u71c3\u4e86\u4eba\u4eec\u5bf9\u5148\u8fdb AI \u7cfb\u7edf\u6f5c\u529b\u7684\u5174\u8da3\u3002\u4f5c\u8005\u6574\u5408\u5e76\u7efc\u8ff0\u4e86 AI \u5bf9\u9f50\u76f8\u5173\u7684\u7814\u7a76\uff0c\u5e76\u5728\u535a\u5ba2\u4e2d\u5c55\u793a\u4e86\u4e0b\u9762\u8fd9\u4e9b\u5bf9\u9f50\u7684\u793a\u4f8b\uff1a<\/span><\/section>\n

<\/span><\/section>\n
<\/section>\n
AI \u5bf9\u9f50\u7684\u793a\u4f8b<\/span><\/section>\n<\/section>\n


<\/span><\/h2>\n


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

\n
\n
\n
\n
\n

<\/section>\n

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

\u80cc\u666f\u9610\u8ff0<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


<\/h3>\n

RICE \u539f\u5219\uff1aAI \u5bf9\u9f50\u7684\u56db\u5927\u652f\u67f1<\/span><\/strong><\/span><\/h3>\n

<\/section>\n
\u5982\u4f55\u6784\u5efa\u7b26\u5408\u4eba\u7c7b\u610f\u56fe\u548c\u4ef7\u503c\u89c2\u7684 AI \u7cfb\u7edf\uff1f<\/span>AI \u5bf9\u9f50\u7684\u6838\u5fc3\u76ee\u6807\u53ef\u4ee5\u6982\u62ec\u4e3a\u56db\u4e2a\u5173\u952e\u539f\u5219\uff1a\u9c81\u68d2\u6027<\/span>\uff08Robustness\uff09<\/span>\u3001\u53ef\u89e3\u91ca\u6027<\/span>\uff08Interpretability\uff09<\/span>\u3001\u53ef\u63a7\u6027<\/span>\uff08Controllability\uff09<\/span>\u548c\u9053\u5fb7\u6027<\/span>\uff08Ethicality\uff09\uff0c<\/span>\u7b80\u79f0\u4e3a RICE<\/span><\/strong>\u3002\u8fd9\u56db\u4e2a\u539f\u5219\u6307\u5bfc\u7740 AI \u7cfb\u7edf\u4e0e\u4eba\u7c7b\u610f\u56fe\u548c\u4ef7\u503c\u89c2\u7684\u4e00\u81f4\u6027\u3002<\/span><\/section>\n

<\/section>\n
<\/p>\n
RICE \u539f\u5219<\/span><\/section>\n

<\/span><\/section>\n
 \u2022 \u9c81\u68d2\u6027\uff1a\u6307 AI \u7cfb\u7edf\u5728\u5404\u79cd\u73af\u5883\u4e0b\u53ef\u9760\u8fd0\u884c\uff0c\u5e76\u80fd\u62b5\u5fa1\u610f\u5916\u5e72\u6270\u7684\u80fd\u529b\u3002<\/span><\/section>\n

<\/section>\n
 \u2022 \u53ef\u89e3\u91ca\u6027\uff1a\u8981\u6c42\u6211\u4eec\u80fd\u591f\u7406\u89e3 AI \u7cfb\u7edf\u5185\u90e8\u7684\u63a8\u7406\u8fc7\u7a0b\uff0c\u7279\u522b\u662f\u4e0d\u900f\u660e\u7684\u795e\u7ecf\u7f51\u7edc\u3002\u901a\u8fc7\u89e3\u91ca\u6027\u5de5\u5177\uff0c\u4f7f\u51b3\u7b56\u8fc7\u7a0b\u5bf9\u7528\u6237\u548c\u5229\u76ca\u76f8\u5173\u8005\u5f00\u653e\u548c\u53ef\u7406\u89e3\uff0c\u4ece\u800c\u786e\u4fdd\u7cfb\u7edf\u7684\u5b89\u5168\u6027\u548c\u53ef\u64cd\u4f5c\u6027\u3002<\/span><\/section>\n

<\/section>\n
 \u2022 \u53ef\u63a7\u6027\uff1a\u786e\u4fdd AI \u7cfb\u7edf\u7684\u884c\u4e3a\u548c\u51b3\u7b56\u8fc7\u7a0b\u53d7\u5230\u4eba\u7c7b\u7684\u76d1\u7763\u548c\u5e72\u9884\u3002\u8fd9\u610f\u5473\u7740\u4eba\u7c7b\u53ef\u4ee5\u53ca\u65f6\u7ea0\u6b63\u7cfb\u7edf\u884c\u4e3a\u4e2d\u7684\u504f\u5dee\uff0c\u786e\u4fdd\u7cfb\u7edf\u5728\u90e8\u7f72\u8fc7\u7a0b\u4e2d\u4fdd\u6301\u5bf9\u9f50\u3002<\/span><\/section>\n

<\/section>\n
 \u2022 \u9053\u5fb7\u6027\uff1aAI \u7cfb\u7edf\u5728\u51b3\u7b56\u548c\u884c\u52a8\u4e2d\u575a\u6301\u793e\u4f1a\u516c\u8ba4\u7684\u9053\u5fb7\u6807\u51c6\uff0c\u5c0a\u91cd\u4eba\u7c7b\u793e\u4f1a\u7684\u4ef7\u503c\u89c2\u3002<\/span><\/section>\n<\/figure>\n


<\/h3>\n

\u5bf9\u9f50\u5faa\u73af<\/span><\/strong><\/h3>\n

<\/span><\/section>\n
\u6211\u4eec\u53ef\u4ee5\u5c06\u5bf9\u9f50\u8fc7\u7a0b\u6784\u5efa\u4e3a\u4e00\u4e2a\u5faa\u73af\uff0c\u53ef\u4ee5\u88ab\u5206\u89e3\u4e3a\u4e24\u4e2a\u5173\u952e\u9636\u6bb5\uff1a\u524d\u5411\u5bf9\u9f50<\/span>\uff08Forward Alignment\uff09<\/span>\u548c\u540e\u5411\u5bf9\u9f50<\/span>\uff08Backward Alignment\uff09<\/span>\u3002<\/span><\/section>\n

<\/span><\/section>\n
<\/p>\n
\u5bf9\u9f50\u5faa\u73af<\/span><\/section>\n

<\/span><\/section>\n
 \u2022 \u524d\u5411\u5bf9\u9f50\u8fc7\u7a0b\uff1a\u4e5f\u79f0\u4f5c\u5bf9\u9f50\u8bad\u7ec3\uff0c\u65e8\u5728\u901a\u8fc7\u5bf9\u9f50\u9700\u6c42\u6765\u8bad\u7ec3\u521d\u6b65\u5bf9\u9f50\u7684\u7cfb\u7edf\uff0c\u4f7f AI \u7cfb\u7edf\u4fdd\u6301\u4e00\u81f4\uff1b<\/span><\/section>\n

<\/section>\n
 \u2022 \u540e\u5411\u5bf9\u9f50\u8fc7\u7a0b\uff1a\u53c8\u79f0\u4f5c\u5bf9\u9f50\u7cbe\u70bc\uff0c\u65e8\u5728\u901a\u8fc7\u7b80\u5355\u548c\u73b0\u5b9e\u73af\u5883\u4e2d\u8fdb\u884c\u8bc4\u4f30\uff0c\u5e76\u8bbe\u7f6e\u76d1\u7ba1\u6761\u4f8b\u548c AI \u6cbb\u7406\uff0c\u4ece\u800c\u786e\u4fdd\u8bad\u7ec3\u7cfb\u7edf\u7684\u5b9e\u9645\u5bf9\u9f50\u3002<\/span><\/section>\n<\/figure>\n

<\/section>\n
\u8fd9\u4e24\u4e2a\u8fc7\u7a0b\u5f62\u6210\u4e00\u4e2a\u5faa\u73af\uff0c\u5176\u4e2d\u524d\u5411\u8fc7\u7a0b\u4e2d\u7684 AI \u7cfb\u7edf\u5bf9\u9f50\u5728\u540e\u5411\u8fc7\u7a0b\u4e2d\u5f97\u5230\u9a8c\u8bc1\uff0c\u540c\u65f6\u4e3a\u4e0b\u4e00\u8f6e\u524d\u5411\u5bf9\u9f50\u63d0\u4f9b\u66f4\u65b0\u7684\u76ee\u6807\u3002<\/span><\/section>\n

<\/section>\n

\u4ece\u53cd\u9988\u4e2d\u5b66\u4e60\u548c\u5728\u5206\u5e03\u504f\u79fb\u4e0b\u5b66\u4e60\u662f AI \u5bf9\u9f50\u4e2d\u7684\u4e24\u4e2a\u91cd\u8981\u65b9\u9762\uff0c\u540c\u65f6\u4e5f\u6784\u6210\u4e86\u524d\u5411\u5bf9\u9f50\u7684\u7ec4\u6210\u90e8\u5206\u3002\u800c\u5bf9\u9f50\u4fdd\u8bc1\u548c AI \u6cbb\u7406\uff0c\u5219\u6784\u6210\u4e86\u540e\u5411\u5bf9\u9f50\u7684\u5143\u7d20\u3002\u5927\u81f4\u6765<\/span>\u8bf4\uff1a<\/span><\/p>\n

 \u2022 \u4ece\u53cd\u9988\u4e2d\u5b66\u4e60\uff1a\u5173\u6ce8\u4e8e\u901a\u8fc7\u4eba\u7c7b\u53cd\u9988\u8bad\u7ec3 AI \u7cfb\u7edf\uff0c\u4f7f\u5176\u4e0e\u4eba\u7c7b\u610f\u56fe\u548c\u4ef7\u503c\u4fdd\u6301\u4e00\u81f4\u3002\u4e3b\u8981\u7814\u7a76\u65b9\u5411\u5305\u62ec\u4ece\u4eba\u7c7b\u53cd\u9988\u4e2d\u5f3a\u5316\u5b66\u4e60<\/span>\uff08RLHF\uff09<\/span>[1][2]\u3001\u504f\u597d\u5efa\u6a21[3]\u548c\u7b56\u7565\u5b66\u4e60\u3002<\/span><\/p>\n

 \u2022 \u5728\u5206\u5e03\u504f\u79fb\u4e0b\u5b66\u4e60\uff1a\u5173\u6ce8\u5728\u8bad\u7ec3\u548c\u90e8\u7f72\u73af\u5883\u4e4b\u95f4\u7684\u5206\u5e03\u53d8\u5316\u4e0b\u4fdd\u6301 AI \u7cfb\u7edf\u7684\u5bf9\u9f50\u3002\u4e3b\u8981\u7814\u7a76\u65b9\u5411\u5305\u62ec\u8de8\u5206\u5e03\u805a\u5408\u3001\u6a21\u5f0f\u8fde\u63a5\u5bfc\u822a\u548c\u5bf9\u6297\u8bad\u7ec3\u3002<\/span><\/p>\n

 \u2022 \u5bf9\u9f50\u4fdd\u8bc1\uff1a\u5173\u6ce8\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u8bc4\u4f30\u548c\u4fdd\u8bc1 AI \u7cfb\u7edf\u7684\u5bf9\u9f50\u3002\u4e3b\u8981\u7814\u7a76\u65b9\u5411\u5305\u62ec\u5b89\u5168\u6d4b\u8bc4\u3001\u53ef\u89e3\u91ca\u6027\u548c\u4eba\u7c7b\u4ef7\u503c\u9a8c\u8bc1\u3002<\/span><\/p>\n

 \u2022 AI \u6cbb\u7406\uff1a\u5173\u6ce8\u4e8e\u5236\u5b9a\u548c\u5b9e\u65bd\u89c4\u5219\uff0c\u786e\u4fdd AI \u7cfb\u7edf\u7684\u5b89\u5168\u53d1\u5c55\u548c\u90e8\u7f72\u3002\u4e3b\u8981\u5305\u62ec\u5f00\u6e90\u6cbb\u7406\u3001\u516c\u5171\u653f\u7b56\u548c\u6cd5\u89c4\u5408\u89c4\u3002<\/span><\/section>\n


<\/h2>\n


<\/p>\n

\n
\n
\n
\n
\n

<\/section>\n

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

\u524d\u5411\u5bf9\u9f50<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


<\/p>\n

\u8fd9\u4e9b\u65b9\u6cd5\u5bf9\u4e8e\u786e\u4fdd AI \u7cfb\u7edf\u5728\u4e0d\u540c\u5206\u5e03\u4e0b\u4fdd\u6301\u4e0e\u4eba\u7c7b\u610f\u56fe\u548c\u4ef7\u503c\u89c2\u4e00\u81f4\u81f3\u5173\u91cd\u8981\u3002<\/span><\/section>\n


<\/h3>\n

\u4ece\u53cd\u9988\u4e2d\u5b66\u4e60<\/span><\/strong><\/span><\/h3>\n

<\/span><\/section>\n
\u4ece\u53cd\u9988\u4e2d\u5b66\u4e60\u662f\u524d\u5411\u5bf9\u9f50\u7684\u8d77\u70b9\uff0c\u6d89\u53ca AI \u7cfb\u7edf\u5982\u4f55\u4ece\u4eba\u7c7b\u53cd\u9988\u4e2d\u5b66\u4e60\uff0c\u5305\u62ec\u4f20\u7edf\u7684\u57fa\u4e8e\u504f\u597d\u7684\u5f3a\u5316\u5b66\u4e60<\/span>\uff08PbRL\uff09<\/span>[4]\u548c\u57fa\u4e8e\u4eba\u7c7b\u53cd\u9988\u7684\u5f3a\u5316\u5b66\u4e60\u3002<\/span><\/section>\n

<\/span><\/section>\n
<\/p>\n
\u4ece\u53cd\u9988\u4e2d\u5b66\u4e60\u7684\u6982\u8ff0<\/span><\/section>\n

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

\u8fd9\u662f\u4e00\u79cd\u901a\u8fc7\u53cd\u9988\u4fe1<\/span>\u606f\u5bf9\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u8fdb\u884c\u8bad\u7ec3\u7684\u65b9\u6cd5\uff0c\u4ee5\u4f7f\u5176\u7b26\u5408\u4eba\u7c7b\u610f\u56fe\u548c\u4ef7\u503c\u3002\u8fd9\u79cd\u5b66\u4e60\u65b9\u6cd5\u5305\u62ec\u591a\u79cd\u8bad\u7ec3\u6280\u672f\uff0c\u5176\u4e2d\u5305\u542b\u4ee5\u4e0b\u51e0\u4e2a\u5143\u7d20\uff1a<\/span><\/p>\n

 \u2022 AI \u7cfb\u7edf\uff1a\u9700\u8981\u5bf9\u9f50\u7684\u5bf9\u8c61\uff0c\u5982\u5bf9\u8bdd\u7cfb\u7edf\u3001\u673a\u5668\u4eba\u7cfb\u7edf\u7b49\u3002<\/span><\/p>\n

 \u2022 \u53cd\u9988<\/span>\uff08Feedback\uff09<\/span>\uff1a\u7531\u987e\u95ee\u96c6\u63d0\u4f9b\uff0c\u8fd9\u4e9b\u53cd\u9988\u7528\u4e8e\u5fae\u8c03 AI \u7cfb\u7edf\u3002<\/span><\/p>\n

 \u2022 \u4ee3\u7406<\/span>\uff08Proxy\uff09<\/span>\uff1a\u7528\u4e8e\u5efa\u6a21\u53cd\u9988\u7684\u7cfb\u7edf\uff0c\u80fd\u4f7f\u7b97\u6cd5\u5b66\u4e60\u66f4\u5bb9\u6613\u8bbf\u95ee\uff0c\u4f8b\u5982\u5728 RLHF \u4e2d\u7684\u5956\u52b1\u6a21\u578b\u3002<\/span><\/section>\n

<\/section>\n

\u4ece\u53cd\u9988\u4e2d\u5b66\u4e60\u8fc7\u7a0b\u4e3b\u8981\u5206\u4e3a\u4e24\u4e2a\u9014\u5f84\uff1a<\/span><\/p>\n

 \u2022 \u76f4\u63a5\u4ece\u53cd\u9988\u4e2d\u5b66\u4e60\uff1aAI \u7cfb\u7edf\u76f4\u63a5\u6839\u636e\u53cd\u9988\u4fe1\u606f\u8fdb\u884c\u5b66\u4e60\u548c\u5fae\u8c03\u3002<\/span><\/p>\n

 \u2022 \u901a\u8fc7\u4ee3\u7406\u8fdb\u884c\u5b66\u4e60\uff1aAI \u7cfb\u7edf\u901a\u8fc7\u5b66\u4e60\u4ee3\u7406\u6a21\u578b<\/span>\uff08\u5982\u5956\u52b1\u6a21\u578b\uff09<\/span>\u6765\u95f4\u63a5\u5730\u4ece\u53cd\u9988\u4fe1\u606f\u4e2d\u5b66\u4e60\u3002<\/span><\/section>\n

<\/section>\n
\u7528\u4e8e\u5bf9\u9f50 AI \u7cfb\u7edf\u7684\u53cd\u9988\u7c7b\u578b\u4e3b\u8981\u6709\u4e09\u79cd\uff1a<\/span><\/section>\n

<\/span><\/section>\n
1. \u5956\u52b1<\/span> (Reward)<\/span> [5]\uff1a\u8fd9\u662f\u5bf9 AI \u7cfb\u7edf\u5355\u4e00\u8f93\u51fa\u7684\u72ec\u7acb\u4e14\u7edd\u5bf9\u7684\u8bc4\u4f30\uff0c\u901a\u5e38\u8868\u793a\u4e3a\u5206\u6570\u3002\u5956\u52b1\u53cd\u9988\u7684\u4f18\u70b9\u662f\u8bbe\u8ba1\u8005\u4e0d\u9700\u8981\u63cf\u8ff0\u6700\u4f18\u884c\u4e3a\uff0c\u540c\u65f6\u5141\u8bb8\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u63a2\u7d22\u4ee5\u627e\u5230\u6700\u4f18\u7b56\u7565\u3002<\/span><\/section>\n
2. \u6f14\u793a<\/span> (Demonstration)<\/span> [6]\uff1a\u8fd9\u662f\u4e13\u5bb6\u987e\u95ee\u5728\u8fbe\u6210\u7279\u5b9a\u76ee\u6807\u65f6\u8bb0\u5f55\u7684\u884c\u4e3a\u6570\u636e\uff0c\u5c55\u793a\u671f\u671b\u7684 AI \u7cfb\u7edf\u884c\u4e3a\uff0c\u4ee5\u4fbf AI \u7cfb\u7edf\u6a21\u4eff\u3002\u8fd9\u79cd\u53cd\u9988\u76f4\u63a5\u5229\u7528\u4e86\u987e\u95ee\u7684\u4e13\u4e1a\u77e5\u8bc6\u548c\u7ecf\u9a8c\uff0c\u65e0\u9700\u5f62\u5f0f\u5316\u7684\u77e5\u8bc6\u8868\u793a\u3002<\/span><\/section>\n
3. \u6bd4\u8f83<\/span> (Comparison)<\/span>\uff1a\u8fd9\u662f\u4e00\u79cd\u76f8\u5bf9\u8bc4\u4f30\uff0c\u5728\u4e00\u7ec4 AI \u7cfb\u7edf\u8f93\u51fa\u8fdb\u884c\u6392\u540d\uff0c\u4ee5\u5f15\u5bfc\u7cfb\u7edf\u505a\u51fa\u66f4\u597d\u7684\u51b3\u7b56\u3002\u8fd9\u79cd\u53cd\u9988\u5f62\u5f0f\u5728\u504f\u597d\u5b66\u4e60\u4e2d\u5f97\u5230\u4f53\u73b0\u3002<\/span><\/section>\n
\u8fd9\u4e9b\u4e0d\u540c\u7c7b\u578b\u7684\u53cd\u9988\u4f53\u73b0\u51fa\u4e00\u4e2a\u5171\u540c\u7279\u5f81\uff1a\u90fd\u53ef\u88ab\u770b\u4f5c\u4eba\u7c7b\u8bd5\u56fe\u4f20\u8fbe\u4e00\u4e2a\u9690\u85cf\u7684\u5956\u52b1\u51fd\u6570\u3002\u90a3\u4e48\uff0c\u5982\u4f55\u4e3a\u66f4\u590d\u6742\u7684\u884c\u4e3a\u5b9a\u4e49\u5956\u52b1\u51fd\u6570\uff0c\u4ee5\u6307\u5bfc\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u7684\u5b66\u4e60\u8fc7\u7a0b\u5462\uff1f<\/span><\/section>\n

<\/span><\/section>\n
\u6211\u4eec\u8fd8\u53ef\u4ee5\u5229\u7528\u504f\u597d\u5efa\u6a21\u3001\u4ece\u4eba\u7c7b\u53cd\u9988\u4e2d\u8fdb\u884c\u5f3a\u5316\u5b66\u4e60<\/span>\uff08RLHF\uff09<\/span>\u548c\u6269\u5c55\u76d1\u7763\u7b49\u65b9\u6cd5\u6765\u66f4\u597d\u5730\u5b9e\u73b0 AI \u7cfb\u7edf\u4e0e\u4eba\u7c7b\u610f\u56fe\u7684\u5bf9\u9f50\u3002<\/span><\/section>\n

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

\u5728\u5206\u5e03\u504f\u79fb\u4e0b\u5b66\u4e60<\/span><\/strong><\/span><\/h3>\n

<\/section>\n
\u53ef\u9760\u7684\u4eba\u5de5\u667a\u80fd\u7cfb\u7edf\u7684\u6784\u5efa\u5728\u5f88\u5927\u7a0b\u5ea6\u4e0a\u4f9d\u8d56\u4e8e\u5b83\u4eec\u9002\u5e94\u591a\u6837\u5316\u6570\u636e\u5206\u5e03\u7684\u80fd\u529b\u3002\u8bad\u7ec3\u6570\u636e\u548c\u5b9e\u9645\u90e8\u7f72\u573a\u666f\u7684\u5206\u5e03\u5f80\u5f80\u5b58\u5728\u7740\u5dee\u5f02\uff0c\u8fd9\u88ab\u79f0\u4e3a\u5206\u5e03\u504f\u79fb[7][8]\u3002<\/span><\/section>\n

<\/span><\/section>\n
\u5206\u5e03\u504f\u79fb\u5e26\u6765\u7684\u4e3b\u8981\u6311\u6218\u662f\u76ee\u6807\u9519\u8bef\u6cdb\u5316[9]\u548c\u81ea\u8bf1\u53d1\u5206\u5e03\u504f\u79fb<\/span>\uff08ADS\uff09<\/span>\u3002\u8fd9\u91cc\u63a2\u8ba8\u4e86\u6570\u636e\u5206\u5e03\u5e72\u9884<\/span>\uff08\u5982\u5bf9\u6297\u8bad\u7ec3\uff09<\/span>\uff0c\u6709\u52a9\u4e8e\u6269\u5c55\u8bad\u7ec3\u6570\u636e\u7684\u5206\u5e03\uff0c\u5e76\u901a\u8fc7\u7b97\u6cd5\u5e72\u9884\u6765\u5bf9\u6297\u76ee\u6807\u9519\u8bef\u6cdb\u5316\u3002<\/span><\/section>\n

<\/span><\/section>\n
<\/p>\n
\u5728\u5206\u5e03\u504f\u79fb\u4e0b\u5b66\u4e60\u7684\u6846\u67b6<\/span><\/section>\n<\/figure>\n

<\/span><\/section>\n
\u5728\u5206\u5e03\u504f\u79fb\u4e0b\u5b66\u4e60\u7684\u6838\u5fc3\u6311\u6218\u662f\u4fdd\u6301 AI \u7cfb\u7edf\u5728\u8bad\u7ec3\u5206\u5e03<\/span>\uff08\u56fa\u5b9a\u8f93\u5165\uff09<\/span>\u548c\u90e8\u7f72\u5206\u5e03<\/span>\uff08\u8f93\u5165\u53d1\u751f\u53d8\u5316\uff09<\/span>\u4e4b\u95f4\u7684\u5bf9\u9f50\u5c5e\u6027<\/span>\uff08\u5373\u9075\u5faa\u4eba\u7c7b\u610f\u56fe\u548c\u4ef7\u503c\uff09<\/span>\u3002\u4ee5\u4e0b\u662f\u4e00\u4e9b\u4e3b\u8981\u65b9\u6cd5\u548c\u6280\u672f\uff1a<\/span><\/section>\n

<\/span><\/section>\n
1.\u8de8\u5206\u5e03\u805a\u5408\uff1a\u5728\u8bad\u7ec3\u673a\u5668\u5b66\u4e60\u6a21\u578b\u65f6\uff0c\u901a\u8fc7\u6700\u5c0f\u5316\u4e0d\u540c\u6570\u636e\u5206\u5e03\u4e4b\u95f4\u7684\u98ce\u9669\uff0c\u6765\u5bfb\u627e\u4e00\u4e2a\u80fd\u591f\u6839\u636e\u56fa\u6709\u5173\u7cfb\u800c\u4e0d\u662f\u5076\u7136\u7279\u5f81\u8fdb\u884c\u9884\u6d4b\u7684\u6a21\u578b\u3002\u8fd9\u79cd\u65b9\u6cd5\u7684\u76ee\u7684\u662f\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u4f18\u5316\u4e0d\u540c\u6570\u636e\u5206\u5e03\u4e0a\u7684\u635f\u5931\u51fd\u6570\uff0c\u4ee5\u51cf\u5c11\u7531\u5206\u5e03\u504f\u79fb\u5f15\u8d77\u7684\u5f71\u54cd\u3002\u5176\u4e2d\u5305\u62ec\u4e00\u4e9b\u65b9\u6cd5\uff0c\u5982\u7ecf\u9a8c\u98ce\u9669\u6700\u5c0f\u5316<\/span>\uff08ERM\uff09<\/span>[10]\u548c\u98ce\u9669\u5916\u63a8<\/span>\uff08REx\uff09<\/span>[11]\u7b49\u3002<\/span><\/section>\n

<\/section>\n
2.\u6a21\u5f0f\u8fde\u63a5\u6307\u5f15\uff1a\u57fa\u4e8e\u6a21\u5f0f\u8fde\u901a\u6027\u8fdb\u884c\u5fae\u8c03\u4ee5\u63d0\u9ad8\u6a21\u578b\u7684\u6cdb\u5316\u6027\u80fd\u3002\u8be5\u65b9\u6cd5\u901a\u8fc7\u57fa\u4e8e\u6a21\u5f0f\u7684\u8fde\u901a\u6027\u6765\u5fae\u8c03\u6a21\u578b\uff0c\u4f7f\u5176\u66f4\u52a0\u5173\u6ce8\u4e0d\u53d8\u7684\u5173\u7cfb\uff0c\u800c\u4e0d\u662f\u5076\u7136\u7684\u76f8\u5173\u6027\u3002\u901a\u8fc7\u6539\u53d8\u5c11\u91cf\u53c2\u6570\uff0c\u5f15\u5bfc\u6a21\u578b\u57fa\u4e8e\u4e0d\u53d8\u5173\u7cfb\u8fdb\u884c\u9884\u6d4b\uff0c\u4ece\u800c\u9f13\u52b1\u6a21\u578b\u6536\u655b\u4e8e\u771f\u5b9e\u76f8\u5173\u6027\u3002\u5305\u62ec\u57fa\u4e8e\u6a21\u5f0f\u8fde\u901a\u7684\u5fae\u8c03<\/span>\uff08CBFT\uff09<\/span>[12]\u65b9\u6cd5\u3002<\/span><\/section>\n

<\/section>\n
3.\u6570\u636e\u5206\u5e03\u5e72\u9884\uff1a\u901a\u8fc7\u5728\u8bad\u7ec3\u6570\u636e\u4e2d\u5f15\u5165\u5bf9\u6297\u6027\u538b\u529b\u548c\u591a\u667a\u80fd\u4f53\u52a8\u6001\uff0c\u4ee5\u6269\u5927\u8bad\u7ec3\u5206\u5e03\uff0c\u7f29\u5c0f\u8bad\u7ec3\u548c\u90e8\u7f72\u5206\u5e03\u4e4b\u95f4\u7684\u5dee\u8ddd\u3002\u5305\u62ec\u5bf9\u6297\u8bad\u7ec3[13][14]\u548c\u5408\u4f5c\u8bad\u7ec3[15]\u7b49\u65b9\u6cd5\u3002<\/span><\/section>\n

<\/section>\n
\u968f\u7740 AI \u7cfb\u7edf\u5728\u73b0\u5b9e\u4e16\u754c\u4e2d\u7684\u5e94\u7528\u8d8a\u6765\u8d8a\u5e7f\u6cdb\uff0c\u6211\u4eec\u9700\u8981\u8003\u8651\u5982<\/span>\u4f55\u5728\u5206\u5e03\u504f\u79fb\u4e0b\u4fdd\u6301 AI \u7cfb\u7edf\u7684\u5bf9\u9f50\u6027\uff0c\u4ee5\u786e\u4fdd\u5b83\u4eec\u5728\u90e8\u7f72\u65f6\u80fd\u591f\u9075\u5faa\u4eba\u7c7b\u610f\u56fe\u548c\u4ef7\u503c\u3002<\/span><\/section>\n

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


<\/h2>\n

\n
\n
\n
\n
\n

<\/section>\n

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

\u540e\u5411\u5bf9\u9f50<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


<\/p>\n

\u540e\u5411\u5bf9\u9f50\u4e3b\u8981\u5305\u62ec\u4e24\u4e2a\u65b9\u9762\uff1a\u8bc4\u4f30<\/span>\uff08Assurance\uff09<\/span>\u548c\u6cbb\u7406<\/span>\uff08Governance\uff09<\/span>\u3002<\/span><\/section>\n

<\/section>\n
\u2022 <\/span>\u8bc4\u4f30\u7684\u76ee\u7684\u662f\u5728 AI \u7cfb\u7edf\u5b9e\u9645\u90e8\u7f72\u540e\u6d4b\u91cf\u548c\u8bc4\u4f30\u5176\u4e0e\u4eba\u7c7b\u610f\u56fe\u7684\u4e00\u81f4\u6027\u3002<\/span> <\/span>
<\/section>\n

<\/section>\n
\u2022 \u6cbb\u7406\u5219\u6d89\u53ca\u521b\u5efa\u548c\u6267\u884c\u786e\u4fdd AI \u7cfb\u7edf\u5b89\u5168\u53d1\u5c55\u548c\u90e8\u7f72\u7684\u89c4\u5219\u3002<\/span><\/section>\n


<\/h3>\n

\u5bf9\u9f50\u4fdd\u8bc1<\/span><\/strong><\/span><\/h3>\n

<\/section>\n
\u5728 AI \u7cfb\u7edf\u7684\u53d1\u5c55\u8fc7\u7a0b\u4e2d\uff0c\u5f53 \u5b9e\u9645\u8bad\u7ec3\u548c\u90e8\u7f72\u4e4b\u540e\uff0c\u786e\u4fdd\u5b83\u4eec\u7684\u884c\u4e3a\u4e0e\u4eba\u7c7b\u610f\u56fe\u548c\u4ef7\u503c\u89c2\u76f8\u5bf9\u9f50\u81f3\u5173\u91cd\u8981\u3002\u4e3a\u4e86\u5b9e\u73b0\u8fd9\u4e00\u76ee\u6807\uff0c\u7814\u7a76\u4eba\u5458\u548c\u5f00\u53d1\u8005\u91c7\u7528\u4e86\u591a\u79cd\u4fdd\u8bc1\u65b9\u6cd5\uff0c\u4ee5\u786e\u4fdd AI \u7cfb\u7edf\u7684\u5b89\u5168\u3001\u53ef\u89e3\u91ca\u6027\u548c\u7b26\u5408\u4eba\u7c7b\u4ef7\u503c\u89c2\u3002<\/span><\/section>\n

<\/span><\/section>\n
<\/p>\n
\u5bf9\u9f50\u4fdd\u8bc1\u7684\u7814\u7a76\u65b9\u5411\u3001\u6280\u672f\u548c\u5e94\u7528<\/span><\/section>\n

<\/span><\/section>\n<\/figure>\n
1. \u5b89\u5168\u6d4b\u8bc4<\/span><\/strong><\/span><\/section>\n

<\/section>\n

\u5b89\u5168\u6d4b\u8bc4\u662f\u786e\u4fdd AI \u7cfb\u7edf\u5bf9\u9f50\u7684\u57fa\u7840\u4fdd\u8bc1\u65b9\u6cd5\uff0c\u5305\u62ec\u5bf9 AI \u7cfb\u7edf\u8fdb\u884c\u5168\u9762\u7684\u98ce\u9669\u8bc4\u4f30\uff0c\u4ee5\u8bc6\u522b\u548c\u7f13\u89e3\u53ef\u80fd\u5bfc\u81f4\u7cfb\u7edf\u884c\u4e3a\u504f\u79bb\u9884\u671f\u76ee\u6807\u7684\u98ce\u9669\u3002\u901a\u5e38\u6d89\u53ca\u4ee5\u4e0b\u51e0\u4e2a\u65b9<\/span>\u9762\uff1a<\/span><\/p>\n

 \u2022 <\/span>\u6570\u636e\u96c6\u548c\u57fa\u51c6\u6d4b\u8bd5\uff1a<\/span>\u901a\u8fc7\u6784\u5efa\u6570\u636e\u96c6\u548c\u57fa\u51c6\u6d4b\u8bd5\u6765\u8bc4\u4f30 AI \u7cfb\u7edf\u7684\u56de\u590d\u3002<\/span>\u8fd9\u4e9b\u6570\u636e\u96c6\u901a\u5e38\u5305\u542b\u9884\u5b9a\u4e49\u7684\u4e0a\u4e0b\u6587\u548c\u4efb\u52a1\uff0c\u4ee5\u5e73\u8861\u6570\u636e\u7684\u6210\u672c\u3001\u8d28\u91cf\u548c\u6570\u91cf\u3002<\/span>
<\/p>\n

 \u2022 \u4ea4\u4e92\u5f0f\u65b9\u6cd5\uff1a\u4e3a\u4e86\u514b\u670d\u6570\u636e\u96c6\u56fa\u6709\u7684\u9759\u6001\u6027\u548c\u53ef\u80fd\u7684\u9488\u5bf9\u6027\u8bad\u7ec3\u95ee\u9898\uff0c\u63d0\u51fa\u4e86\u65b0\u7684\u4ea4\u4e92\u5f0f\u4fdd\u8bc1\u65b9\u6cd5\uff0c\u5305\u62ec\u4f7f\u7528\u667a\u80fd\u4f53<\/span>\uff08\u5982\u4eba\u7c7b\u6216\u66f4\u9ad8\u7ea7\u7684\u8bed\u8a00\u6a21\u578b\uff09<\/span>\u6765\u8bc4\u4f30 AI \u6a21\u578b\u7684\u8f93\u51fa\uff0c\u4ee5\u53ca\u901a\u8fc7\u6a21\u62df\u73af\u5883\u4ea4\u4e92\u6765\u8bc4\u4f30 AI \u7cfb\u7edf\u7684\u5bf9\u9f50\u8d28\u91cf\u3002<\/span><\/section>\n

<\/section>\n
2. \u53ef\u89e3\u91ca\u6027<\/span><\/strong><\/span><\/section>\n

<\/section>\n

\u53ef\u89e3\u91ca\u6027\u662f\u6307\u8ba9\u4eba\u7c7b\u80fd\u591f\u7406\u89e3\u673a\u5668\u5b66\u4e60\u7cfb\u7edf\u53ca\u5176\u51b3\u7b56\u8fc7\u7a0b\u7684\u7814\u7a76\u9886\u57df\u3002<\/span><\/p>\n

 \u2022 \u901a\u8def\u5206\u6790[16]\uff1a\u7814\u7a76\u795e\u7ecf\u7f51\u7edc\u5185\u90e8\u7684\u5b50\u7f51\u7edc<\/span>\uff08\u73af\u8def\uff09<\/span>\uff0c\u8fd9\u4e9b\u5b50\u7f51\u7edc\u5177\u6709\u7279\u5b9a\u7684\u529f\u80fd\u3002\u7814\u7a76\u4eba\u5458\u5728\u795e\u7ecf\u7f51\u7edc\u4e2d\u5b9a\u4f4d\u73af\u8def<\/span>\uff08\u5fae\u89c2\uff09<\/span>\u53ef\u4ee5\u7406\u89e3\u6a21\u578b\u884c\u4e3a(\u5b8f\u89c2)\u3002<\/span><\/p>\n

 \u2022 \u5f52\u56e0\u5206\u6790\uff1a\u5f52\u56e0\u6280\u672f\u8bc4\u4f30\u6a21\u578b\u7ec4\u4ef6<\/span>\uff08\u5982\u5f52\u7eb3\u5934\u3001\u795e\u7ecf\u5143\u3001\u5c42\u548c\u8f93\u5165\uff09<\/span>\u5bf9\u795e\u7ecf\u5143\u54cd\u5e94\u548c\u6a21\u578b\u8f93\u51fa\u7684\u8d21\u732e\u3002<\/span><\/p>\n

 \u2022 \u53ef\u89c6\u5316\uff1a\u53ef\u89c6\u5316\u6280\u672f\u5e2e\u52a9\u7406\u89e3\u795e\u7ecf\u7ed3\u6784\uff0c\u5305\u62ec\u6570\u636e\u96c6\u3001\u7279\u5f81\u3001\u6743\u91cd\u3001\u6fc0\u6d3b\u548c\u6574\u4e2a\u795e\u7ecf\u7f51\u7edc\u7684\u53ef\u89c6\u5316\u3002<\/span><\/p>\n

 \u2022 \u6270\u52a8\u548c\u6d88\u878d\uff1a\u65e8\u5728\u6d4b\u8bd5\u6a21\u578b\u63a8\u7406\u7684\u53cd\u4e8b\u5b9e\u6027\u800c\u975e\u76f8\u5173\u6027\u3002\u8fd9\u4e9b\u6280\u672f\u6709\u52a9\u4e8e\u5efa\u7acb\u795e\u7ecf\u6fc0\u6d3b\u4e0e\u6574\u4e2a\u7f51\u7edc\u884c\u4e3a\u4e4b\u95f4\u7684\u56e0\u679c\u5173\u7cfb\u3002<\/span><\/p>\n

 \u2022 \u6620\u5c04\u548c\u7f16\u8f91\u5b66\u4e60\u5230\u7684\u8868\u793a\uff1a\u4e0e\u6a21\u578b\u8f93\u51fa\u7684\u5185\u5bb9\u76f8\u6bd4\uff0c\u77e5\u8bc6\u8868\u793a\u6620\u5c04\u548c\u7f16\u8f91\u6280\u672f\u6709\u52a9\u4e8e\u7406\u89e3\u5927\u8bed\u8a00\u6a21\u578b\u771f\u6b63\u638c\u63e1\u7684\u5185\u5bb9\uff0c\u5e76\u5728\u8fd9\u4e9b\u77e5\u8bc6\u4e0d\u771f\u5b9e\u65f6\u4fee\u6539\u5927\u8bed\u8a00\u6a21\u578b\u7684\u77e5\u8bc6\u8868\u793a\u3002\u8fd9\u4e9b\u6280\u672f\u80fd\u591f\u76f4\u89c2\u5730\u5e2e\u52a9\u68c0\u6d4b\u6b3a\u9a97\u3002<\/span><\/section>\n

<\/section>\n
3. \u4eba\u7c7b\u4ef7\u503c\u5951\u5408\u6027\u9a8c\u8bc1<\/span><\/strong><\/span><\/section>\n

<\/section>\n

\u8981\u786e\u4fdd AI \u7cfb\u7edf\u5728\u6267\u884c\u4efb\u52a1\u6216\u534f\u52a9\u4eba\u7c7b\u51b3\u7b56\u65f6\u9075\u5faa\u4eba\u7c7b\u7684\u793e\u4f1a\u548c\u9053\u5fb7\u89c4\u8303\uff0c\u8fd9\u662f AI \u80fd\u878d\u5165\u4eba\u7c7b\u793e\u4f1a\u7684\u9ad8\u7ea7\u9700\u6c42\u3002\u8bc4\u4f30\u65b9\u6cd5\u5305\u62ec\uff1a<\/span><\/p>\n

 \u2022 \u6784\u5efa\u9053\u5fb7\u6570\u636e\u96c6\uff1a\u901a\u8fc7\u6784\u5efa\u6570\u636e\u96c6\u6765\u8bc4\u4f30 AI \u7cfb\u7edf\u662f\u5426\u7b26\u5408\u4eba\u7c7b\u7684\u793e\u4f1a\u548c\u9053\u5fb7\u89c4\u8303\u3002\u4f8b\u5982\uff0cSOCIALCHEM-101 \u6570\u636e\u96c6[17]\u63d0\u4f9b\u4e86\u4eba\u7c7b\u793e\u4f1a\u548c\u9053\u5fb7\u89c4\u8303\u7684\u6307\u5bfc\u3002<\/span><\/p>\n

 \u2022 \u573a\u666f\u6a21\u62df\uff1a\u901a\u8fc7\u6587\u672c\u5192\u9669\u6e38\u620f\u7b49\u573a\u666f\u6a21\u62df\u6765\u8bc4\u4f30 AI \u7cfb\u7edf\u5728\u590d\u6742\u884c\u4e3a\u4e2d\u7684\u9053\u5fb7\u8868\u73b0\uff0c\u5982\u6b3a\u9a97\u3001\u64cd\u7eb5\u548c\u80cc\u53db\u3002<\/span><\/p>\n

 \u2022 \u4ef7\u503c\u8bc4\u4f30\u65b9\u6cd5\uff1a\u901a\u8fc7\u6bd4\u8f83 AI \u7cfb\u7edf\u4ea7\u751f\u7684\u56de\u590d\u4e0e\u4e0d\u540c\u6587\u5316\u7fa4\u4f53\u7684\u4ef7\u503c\u89c2\u4e4b\u95f4\u7684\u76f8\u4f3c\u6027\u6765\u8bc4\u4f30 AI \u7cfb\u7edf\u7684\u4ef7\u503c\u53d6\u5411\u3002<\/span><\/section>\n


<\/h3>\n

AI \u6cbb\u7406<\/span><\/strong><\/span><\/h3>\n

<\/section>\n
AI \u6cbb\u7406\u662f\u901a\u8fc7\u653f\u7b56\u548c\u89c4\u5219\u6765\u786e\u4fdd AI \u7cfb\u7edf\u7684\u5b89\u5168\u5f00\u53d1\u548c\u90e8\u7f72\u3002\u5728 AI \u6cbb\u7406\u4e2d\uff0c\u591a\u65b9\u5229\u76ca\u76f8\u5173\u8005\u53d1\u6325\u7740\u91cd\u8981\u4f5c\u7528\uff0c\u5305\u62ec\u653f\u5e9c\u673a\u6784\u3001\u884c\u4e1a\u548c AGI \u5b9e\u9a8c\u5ba4\u4ee5\u53ca\u7b2c\u4e09\u65b9\u7ec4\u7ec7\u3002<\/span><\/section>\n

<\/span><\/section>\n
<\/p>\n
\u5206\u6790 AI \u6cbb\u7406\u7684\u6846\u67b6<\/span><\/section>\n

<\/section>\n
1. \u653f\u5e9c\u673a\u6784\uff1a\u901a\u8fc7\u7acb\u6cd5\u3001\u53f8\u6cd5\u548c\u6267\u6cd5\u6743\u529b\u76d1\u7763AI\u653f\u7b56\uff0c\u5e76\u53c2\u4e0e\u56fd\u9645\u5408\u4f5c\u3002\u5b83\u4eec\u5236\u5b9a AI \u5b89\u5168\u6807\u51c6\uff0c\u8981\u6c42 AI \u5f00\u53d1\u8005\u6ce8\u518c\u548c\u62a5\u544a\uff0c\u4ee5\u53ca\u786e\u4fdd\u9075\u5b88\u5b89\u5168\u6807\u51c6\u3002<\/span><\/section>\n

<\/section>\n
2. \u884c\u4e1a\u548c AGI \u5b9e\u9a8c\u5ba4\uff1a\u884c\u4e1a\u548c AGI \u5b9e\u9a8c\u5ba4\u7814\u7a76\u548c\u90e8\u7f72 AI \u6280\u672f\uff0c\u540c\u65f6\u63d0\u51fa\u81ea\u6211\u6cbb\u7406\u7684\u6280\u672f\u548c\u5f71\u54cd\u6cbb\u7406\u653f\u7b56\u3002\u5b83\u4eec\u6267\u884c\u5168\u9762\u7684 AI \u98ce\u9669\u8bc4\u4f30\uff0c\u5e76\u91c7\u53d6\u5b89\u5168\u63aa\u65bd\uff0c\u5982\u7b2c\u4e09\u65b9\u6a21\u578b\u5ba1\u8ba1\u548c\u7ea2\u961f\u6d4b\u8bd5\u3002<\/span><\/section>\n

<\/section>\n
3. \u7b2c\u4e09\u65b9\u7ec4\u7ec7\uff1a\u7b2c\u4e09\u65b9\u7ec4\u7ec7<\/span>\uff08\u5305\u62ec\u5b66\u672f\u754c\u3001\u975e\u653f\u5e9c\u7ec4\u7ec7\u548c\u975e\u8425\u5229\u7ec4\u7ec7\uff09<\/span>\uff0c\u4e0d\u4ec5\u5bf9\u516c\u53f8\u6cbb\u7406\u3001AI \u7cfb\u7edf\u53ca\u5176\u5e94\u7528\u8fdb\u884c\u5ba1\u8ba1\uff0c\u8fd8\u534f\u52a9\u653f\u5e9c\u5236\u5b9a\u653f\u7b56\u3002<\/span><\/section>\n<\/figure>\n


<\/span><\/h3>\n

AI \u5bf9\u9f50\u9762\u4e34\u7684\u5168\u7403\u6027\u95ee\u9898\u548c\u5f00\u653e\u6027\u6311\u6218<\/span><\/strong><\/span><\/h3>\n


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

\u968f\u7740 AI \u6280\u672f\u7684\u5168\u7403\u666e\u53ca\u548c\u5f71\u54cd\u529b\u7684\u589e\u52a0\uff0c\u56fd\u9645\u6cbb\u7406\u548c\u5f00\u6e90\u6cbb\u7406\u6210\u4e3a AI \u5bf9\u9f50\u4e2d\u7684\u91cd\u8981\u8bae\u9898\u3002\u8fd9\u4e9b\u95ee\u9898\u5f80\u5f80\u6ca1\u6709\u660e\u786e\u7684\u7b54\u6848\uff0c\u800c\u5bf9\u95ee\u9898\u7684\u8ba8\u8bba\u5f80\u5f80\u53ef\u4ee5\u4fc3\u8fdb\u66f4\u597d\u7684\u6cbb\u7406\u3002<\/span><\/section>\n

<\/span><\/section>\n
 \u2022 \u56fd\u9645\u6cbb\u7406\uff1a\u56fd\u9645\u6cbb\u7406\u7684\u76ee\u6807\u662f\u901a\u8fc7\u5168\u7403\u6846\u67b6\u6765\u964d\u4f4e AI \u7cfb\u7edf\u5e26\u6765\u7684\u98ce\u9669\uff0c\u5e76\u786e\u4fdd AI \u53d1\u5c55\u7684\u673a\u4f1a\u516c\u5e73\u5206\u914d\u3002\u76ee\u524d\uff0c\u591a\u4e2a\u56fd\u9645\u7ec4\u7ec7\u5df2\u5c31 AI \u6cbb\u7406\u8fbe\u6210\u5171\u8bc6\uff0c\u5982 G20 \u548c OECD\u3002<\/span><\/section>\n

<\/section>\n
 \u2022 \u5f00\u6e90\u6cbb\u7406\uff1a\u5f00\u6e90\u6cbb\u7406\u7684\u8fa9\u8bba\u96c6\u4e2d\u5728\u662f\u5426\u5e94\u8be5\u516c\u5f00\u6700\u65b0\u7684 AI \u6a21\u578b\u3002\u652f\u6301\u8005\u8ba4\u4e3a\u8fd9\u53ef\u4ee5\u589e\u5f3a\u6a21\u578b\u7684\u5b89\u5168\u6027\uff0c\u4fc3\u8fdb\u6743\u529b\u548c\u63a7\u5236\u7684\u53bb\u4e2d\u5fc3\u5316\u3002\u800c\u53cd\u5bf9 \u8005\u5219\u62c5\u5fc3\u5f00\u6e90\u6a21\u578b\u53ef\u80fd\u88ab\u6ee5\u7528\uff0c\u5bfc\u81f4\u5b89\u5168\u98ce\u9669\u3002<\/span><\/section>\n


<\/span><\/h2>\n


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

\n
\n
\n
\n
\n

<\/section>\n

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

\u603b\u7ed3<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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


<\/p>\n

\u8fd1\u51e0\u5e74\u6765\uff0cAI \u5bf9\u9f50\u9886\u57df\u7684\u591a\u6837\u6027\u6709\u4e86\u521b\u65b0\uff0c\u4f7f\u4e0d\u540c\u7684\u7814\u7a76\u65b9\u5411\u7684\u76f8\u4e92\u7ade\u4e89\u548c\u78b0\u649e\uff0c\u4fc3\u8fdb\u4e86\u601d\u60f3\u7684\u4ea4\u6d41\u4f20\u64ad\u3002\u540c\u65f6\uff0c\u8fd9\u79cd\u591a\u6837\u6027\u4e5f\u63d0\u9ad8\u4e86\u8fdb\u5165\u8be5\u9886\u57df\u7684\u95e8\u69db\u3002<\/span><\/section>\n

<\/span><\/section>\n
\u672c\u6587\u5bf9 AI \u5bf9\u9f50\u9886\u57df\u8fdb\u884c\u7efc\u8ff0\uff0c\u5305\u62ec\u524d\u5411\u5bf9\u9f50\u548c\u540e\u5411\u5bf9\u9f50\u7684\u7814\u7a76\u65b9\u5411\uff0c\u4ee5\u53ca\u4fdd\u8bc1\u65b9\u6cd5\u548c\u6cbb\u7406\u5b9e\u8df5\u3002AI \u5bf9\u9f50\u9886\u57df\u7684\u591a\u6837\u6027\u5e26\u6765\u4e86\u521b\u65b0\u548c\u6311\u6218\uff0c\u8981\u6c42\u6211\u4eec\u91c7\u53d6\u5f00\u653e\u5f0f\u63a2\u7d22\u548c\u591a\u89d2\u5ea6\u8003\u8651\uff0c\u4ee5\u786e\u4fdd AI \u7cfb\u7edf\u7684\u5b89\u5168\u3001\u53ef\u89e3\u91ca\u6027\u548c\u4eba\u7c7b\u4ef7\u503c\u89c2\u5951\u5408\u6027\u3002<\/span><\/section>\n

<\/span><\/section>\n
AI \u5bf9\u9f50\u7684\u7814\u7a76\u9700\u8981\u7ed3\u5408\u524d\u77bb\u6027\u548c\u9762\u5411\u5f53\u4e0b\u7684\u89c6\u89d2\uff0c\u672a\u6765\u7684\u7814\u7a76\u9700\u8981\u6301\u7eed\u66f4\u65b0\uff0c\u4ee5\u53cd\u6620\u673a\u5668\u5b66\u4e60\u7684\u6700\u65b0\u53d1\u5c55\uff0c\u5e76\u5f3a\u8c03\u653f\u7b56\u76f8\u5173\u6027\u548c\u793e\u4f1a\u590d\u6742\u6027\u5728 AI \u5bf9\u9f50\u4e2d\u7684\u91cd\u8981\u7a0b\u5ea6\u3002<\/span><\/section>\n

<\/span><\/section>\n
\u968f\u7740 AI \u7cfb\u7edf\u8d8a\u6765\u8d8a\u591a\u5730\u878d\u5165\u793e\u4f1a\uff0c\u5bf9\u9f50\u5c06\u4e0d\u518d\u4ec5\u662f\u5355\u7eaf\u7684\u667a\u80fd\u4f53\u95ee\u9898\uff0c\u800c\u662f\u4e00\u4e2a\u793e\u4f1a\u95ee\u9898\u3002\u8fd9\u8981\u6c42 AI \u7cfb\u7edf\u4e0d\u4ec5\u8981\u4e0e\u4eba\u7c7b\u610f\u56fe\u5bf9\u9f50\uff0c\u8fd8\u8981\u8003\u8651\u793e\u4f1a\u590d\u6742\u6027\u548c\u9053\u5fb7\u4ef7\u503c\u3002<\/span><\/section>\n

<\/section>\n

<\/section>\n

\u53c2\u8003\u8d44\u6599<\/span><\/strong><\/h4>\n

[1]CHRISTIANO P F, LEIKE J, BROWN T, et al. Deep reinforcement learning from human preferences[J]. Advances in neural information processing systems, 2017, 30.<\/span><\/p>\n

[2]BAI Y, JONES A, NDOUSSE K, et al. Training a helpful and harmless assistant with reinforcement learning from human feedback[A]. 2022.<\/span><\/p>\n

[3]AKROUR R, SCHOENAUER M, SEBAG M. Preference-based policy learning[C]\/\/Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2011, Athens, Greece, September 5-9, 2011. Proceedings, Part I 11. Springer, 2011: 12-27.<\/span><\/p>\n

[4]WIRTH C, AKROUR R, NEUMANN G, et al. A survey of preference-based reinforcement learning methods[J]. Journal of Machine Learning Research, 2017, 18(136): 1-46.<\/span><\/p>\n

[5]SILVER D, SINGH S, PRECUP D, et al. Reward is enough[J]. Artificial Intelligence, 2021, 299: 103535<\/span><\/p>\n

[6]HUSSEIN A, GABER M M, ELYAN E, et al. Imitation learning: A survey of learning methods[J]. ACM Computing Surveys (CSUR), 2017, 50(2): 1-35.<\/span><\/p>\n

[7]KRUEGER D, MAHARAJ T, LEIKE J. Hidden incentives for auto-induced distributional shift[A]. 2020.<\/span><\/p>\n

[8]THULASIDASAN S, THAPA S, DHAUBHADEL S, et al. An effective baseline for robustness to distributional shift[C]\/\/2021 20th IEEE International Conference on Machine Learning and Applications (ICMLA). IEEE, 2021: 278-285.<\/span><\/p>\n

[9]SHAH R, VARMA V, KUMAR R, et al. Goal misgeneralization: Why correct specifications aren\u2019t enough for correct goals[A]. 2022.<\/span><\/p>\n

[10]VAPNIK V. Principles of risk minimization for learning theory[J]. Advances in neural information processing systems, 1991, 4.<\/span><\/p>\n

[11]KRUEGER D, CABALLERO E, JACOBSEN J H, et al. Out-of-distribution generalization via risk extrapolation (rex)[C]\/\/International Conference on Machine Learning. PMLR, 2021: 5815-5826.<\/span><\/p>\n

[12]LUBANA E S, BIGELOW E J, DICK R P, et al. Mechanistic mode connectivity[C]\/\/International Conference on Machine Learning. PMLR, 2023: 22965-23004.<\/span><\/p>\n

[13]SONG Y, SHU R, KUSHMAN N, et al. Constructing unrestricted adversarial examples with generative models[J]. Advances in Neural Information Processing Systems, 2018, 31.<\/span><\/p>\n

[14]YOO J Y, QI Y. Towards improving adversarial training of NLP models[C\/OL]\/\/Findings of the Association for Computational Linguistics: EMNLP 2021. Punta Cana, Dominican Republic: Association for Computational Linguistics, 2021: 945-956. https:\/\/aclanthology.org\/2021.findings-emnlp.81. DOI: 10.18653\/v1\/2021.findings-emnlp.81.<\/span>
<\/p>\n

[15]DAFOE A, HUGHES E, BACHRACH Y, et al. Open problems in cooperative ai[A]. 2020.<\/span><\/p>\n

[16]R\u00c4UKER T, HO A, CASPER S, et al. Toward transparent ai: A survey on interpreting the inner structures of deep neural networks[C]\/\/2023 IEEE Conference on Secure and Trustworthy Machine Learning (SaTML). IEEE, 2023: 464-483.<\/span>
<\/p>\n

[17]FORBES M, HWANG J D, SHWARTZ V, et al. Social chemistry 101: Learning to reason about social and moral norms[C\/OL]\/\/WEBBER B, COHN T, HE Y, et al. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, EMNLP 2020, Online, November 16-20, 2020. Association for Computational Linguistics, 2020: 653-670. https:\/\/doi.org\/10.18653\/v1\/2020.emnlp-main.48. DOI: 10.18653\/V1\/2020.EMNLP-MAIN.48.<\/span><\/p>\n

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

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

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

\u5927\u6a21\u578b\u5b89\u5168\u4e0e\u5bf9\u9f50\u8bfb\u4e66\u4f1a<\/strong>
<\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n


<\/section>\n

\u5927\u6a21\u578b\u7684\u72c2\u98d9\u7a81\u8fdb\u5524\u9192\u4e86\u4eba\u4eec\u5bf9AI\u6280\u672f\u7684\u70ed\u60c5\u548c\u61a7\u61ac\uff0c\u4e5f\u5f15\u53d1\u4e86\u5bf9AI\u6280\u672f\u672c\u8eab\u5b58\u5728\u7684\u793e\u4f1a\u4f26\u7406\u98ce\u9669\u53ca\u5176\u5bf9\u4eba\u7c7b\u751f\u5b58\u6784\u6210\u7684\u6f5c\u5728\u5a01\u80c1\u7684\u666e\u904d\u62c5\u5fe7\u3002\u5728\u6b64\u80cc\u666f\u4e0b\uff0cAI\u5b89\u5168\u4e0e\u5bf9\u9f50\u5f97\u5230\u5e7f\u6cdb\u5173\u6ce8\uff0c\u8fd9\u662f\u4e00\u4e2a\u81f4\u529b\u4e8e\u8ba9AI\u9020\u798f\u4eba\u7c7b\uff0c\u907f\u514dAI\u6a21\u578b\u5931\u63a7\u6216\u88ab\u6ee5\u7528\u800c\u5bfc\u81f4\u707e\u96be\u6027\u540e\u679c\u7684\u7814\u7a76\u65b9\u5411\u3002\u96c6\u667a\u4ff1\u4e50\u90e8\u548c\u5b89\u8fdcAI\u8054\u5408\u4e3e\u529e\u300c\u5927\u6a21\u578b\u5b89\u5168\u4e0e\u5bf9\u9f50\u300d\u8bfb\u4e66\u4f1a<\/a>\uff0c\u7531\u591a\u4f4d\u6d77\u5185\u5916\u4e00\u7ebf\u7814\u7a76\u8005\u8054\u5408\u53d1\u8d77\uff0c\u65e8\u5728\u6df1\u5165\u63a2\u8ba8AI\u5b89\u5168\u4e0e\u5bf9\u9f50\u6240\u6d89\u53ca\u7684\u6838\u5fc3\u6280\u672f\u3001\u7406\u8bba\u67b6\u6784\u3001\u89e3\u51b3\u8def\u5f84\u4ee5\u53ca\u5b89\u5168\u6cbb\u7406\u7b49\u4ea4\u53c9\u8bfe\u9898\u3002<\/span><\/p>\n


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

<\/section>\n
\u8be6\u60c5\u8bf7\u89c1\uff1a<\/span><\/section>\n
2024\u5f00\u5e74\u8bfb\u4e66\u4f1a\uff1aAI\u5b89\u5168\u4e0e\u5bf9\u9f50\u2014\u2014\u5e94\u5bf9\u524d\u6cbfAI\u5931\u63a7\u4e0e\u6ee5\u7528\u7684\u6280\u672f\u8def\u7ebf<\/a><\/span>
<\/span><\/section>\n

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


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

\n

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

1. <\/strong>AI\u5bf9\u9f50\u662f\u63a7\u5236\u8bba\u8fd8\u662f\u535a\u5f08\u8bba\uff1f<\/a><\/strong><\/span><\/strong><\/strong><\/section>\n
2.<\/strong> Yoshua Bengio\u6f14\u8bb2\u3001NeurIPS\u5bf9\u9f50\u5de5\u4f5c\u574a\uff1a\u907f\u514dAI\u707e\u96be\u6027\u98ce\u9669\u7684\u6280\u672f\u4e0e\u534f\u8c03\u6311\u6218<\/a><\/strong><\/span><\/section>\n
3. <\/strong><\/strong>\u5927\u6a21\u578b\u5b89\u5168\u4e0e\u5bf9\u9f50\uff1a\u590d\u6742\u7cfb\u7edf\u89c6\u89d2\u4e0b\u7684AI\u5b89\u5168<\/a><\/strong><\/strong><\/span><\/strong><\/strong><\/section>\n
4. <\/strong><\/strong><\/span>\u5f20\u6c5f\uff1a\u7b2c\u4e09\u4ee3\u4eba\u5de5\u667a\u80fd\u6280\u672f\u57fa\u7840\u2014\u2014\u4ece\u53ef\u5fae\u5206\u7f16\u7a0b\u5230\u56e0\u679c\u63a8\u7406 | \u96c6\u667a\u5b66\u56ed\u5168\u65b0\u8bfe\u7a0b<\/a><\/strong><\/strong><\/span><\/section>\n
<\/a><\/strong><\/strong><\/span><\/section>\n
5.<\/strong><\/strong> \u9f99\u5e74\u5927\u8fd0\u8d77\uff0c\u5b66\u4e60\u6b63\u5f53\u65f6\uff01\u89e3\u9501\u96c6\u667a\u5168\u7ad9\u5185\u5bb9\uff0c\u5f00\u542f\u65b0\u5e74\u5b66\u4e60\u8ba1\u5212<\/a><\/strong><\/strong><\/strong><\/strong><\/span><\/section>\n
6. <\/strong>\u52a0\u5165\u96c6\u667a\uff0c\u4e00\u8d77\u590d\u6742\uff01<\/strong><\/span><\/a><\/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 \u8fd1\u671f\uff0c\u6765\u81ea\u5317\u5927\u7684\u7814\u7a76\u56e2\u961f\u6574\u7406\u4e86\u4e00\u4efd\u8be6\u5c3d\u7684 AI \u5bf9\u9f50\u6700\u65b0\u7efc\u8ff0\uff0c\u672c\u6587\u6982\u8ff0\u4e86\u5f53\u524d\u4eba\u5de5\u667a\u80fd\u5bf9\u9f50\u7814\u7a76\u7684\u5168\u8c8c\uff0c\u4f9d\u636e\u56db\u4e2a\u5173\u952e\u539f\u5219\uff0c\u5c06\u5176\u5206\u89e3\u4e3a\u4e24\u4e2a\u5173\u952e\u7ec4\u6210\u90e8\u5206\uff1a\u524d\u5411\u5bf9\u9f50\u548c\u540e\u5411\u5bf9\u9f50\uff0c\u4ece\u800c\u8fdb\u884c\u66f4\u5168\u9762\u7684\u8ba8\u8bba\u3002\u5f53\u524d\u7684\u7814\u7a76\u548c\u5b9e\u8df5\u5c06\u8fd9\u4e9b\u76ee\u6807\u7eb3\u5165\u4e86\u53cd\u9988\u5b66\u4e60\u3001\u5206\u5e03\u504f\u79fb\u5b66\u4e60\u3001\u4fdd\u8bc1\u548c\u6cbb\u7406\u7b49\u56db\u4e2a\u9886\u57df\u3002AI \u5bf9\u9f50\u7684\u76ee\u6807\u4e0d\u4ec5\u4ec5\u662f\u4e3a\u4e86…<\/p>\n","protected":false},"author":0,"featured_media":48529,"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\/48540"}],"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=48540"}],"version-history":[{"count":0,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/48540\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/48529"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=48540"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=48540"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=48540"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=48540"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}