{"id":42390,"date":"2023-05-08T16:43:09","date_gmt":"2023-05-08T08:43:09","guid":{"rendered":"https:\/\/swarma.org\/?p=42390"},"modified":"2023-05-08T16:43:09","modified_gmt":"2023-05-08T08:43:09","slug":"%e5%9b%a0%e6%9e%9c%e6%8e%a8%e7%90%86%e4%b8%8e%e5%a4%a7%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8b%ef%bc%9a%e5%bc%80%e8%be%9f%e5%9b%a0%e6%9e%9c%e5%85%b3%e7%b3%bb%e7%9a%84%e6%96%b0%e5%89%8d%e6%b2%bf","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=42390","title":{"rendered":"\u56e0\u679c\u63a8\u7406\u4e0e\u5927\u8bed\u8a00\u6a21\u578b\uff1a\u5f00\u8f9f\u56e0\u679c\u5173\u7cfb\u7684\u65b0\u524d\u6cbf"},"content":{"rendered":"
\n
\n
\n
\n
<\/div>\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

\u5927\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u7684\u56e0\u679c\u63a8\u7406\u80fd\u529b\u4e00\u76f4\u662f\u4e00\u4e2a\u4e89\u8bae\u6027\u7684\u95ee\u9898\uff0c\u8981\u5728\u533b\u5b66\u3001\u79d1\u5b66\u3001\u6cd5\u5f8b\u548c\u653f\u7b56\u7b49\u5bf9\u793e\u4f1a\u4ea7\u751f\u91cd\u5927\u5f71\u54cd\u7684\u9886\u57df\u4f7f\u7528\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u7814\u7a76\u5176\u56e0\u679c\u63a8\u7406\u80fd\u529b\u5177\u6709\u91cd\u8981\u610f\u4e49\u3002\u8fd1\u65e5\u53d1\u8868\u4e8e arXiv \u7684\u6700\u65b0\u8bba\u6587\u201c\u56e0\u679c\u63a8\u7406\u4e0e\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff1a\u5f00\u8f9f\u56e0\u679c\u5173\u7cfb\u7684\u65b0\u524d\u6cbf\u201d\uff0c\u6df1\u5165\u63a2\u8ba8\u5927\u8bed\u8a00\u6a21\u578b\u53ca\u5176\u56e0\u679c\u63a8\u7406\u80fd\u529b\u3002<\/span><\/strong><\/p>\n


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

\u5728\u8fd9\u9879\u7814\u7a76\u4e2d\uff0c\u57fa\u4e8e\u5927\u8bed\u8a00\u6a21\u578b\u7684\u65b9\u6cd5\u5728\u56e0\u679c\u53d1\u73b0\u3001\u53cd\u4e8b\u5b9e\u63a8\u7406\u548c\u5b9e\u9645\u56e0\u679c\u5173\u7cfb\u7b49\u591a\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u6700\u9ad8\u7684\u51c6\u786e\u6027\u3002\u901a\u8fc7\u6355\u6349\u5173\u4e8e\u56e0\u679c\u673a\u5236\u7684\u5e38\u8bc6\u548c\u9886\u57df\u77e5\u8bc6\uff0c\u5e76\u652f\u6301\u81ea\u7136\u8bed\u8a00\u4e0e\u5f62\u5f0f\u65b9\u6cd5\u4e4b\u95f4\u7684\u8f6c\u6362\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u4e3a\u63a8\u8fdb\u56e0\u679c\u63a8\u7406\u5f00\u8f9f\u4e86\u65b0\u524d\u6cbf\u3002<\/span><\/strong><\/p>\n<\/section>\n

\n

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

\u5173\u952e\u8bcd\uff1a\u5927\u8bed\u8a00\u6a21\u578b\uff0c\u56e0\u679c\u63a8\u7406<\/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
<\/strong><\/strong><\/strong><\/section>\n
<\/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

\u6881\u91d1<\/span> <\/span>| \u7f16\u8bd1<\/strong><\/span><\/span><\/p>\n


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


<\/section>\n
\n
\u8bba\u6587\u9898\u76ee\uff1aCausal Reasoning and Large Language Models: Opening a New Frontier for Causality<\/span><\/section>\n
\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/2305.00050<\/span><\/section>\n
\u4f5c\u8005\uff1aE K\u0131c\u0131man, R Ness, A Sharma, C Tan [Microsoft Research & University of Chicago]<\/span><\/section>\n<\/blockquote>\n

<\/section>\n
\u8bfa\u8d1d\u5c14\u7269\u7406\u5b66\u5956\u5f97\u4e3b\u5c24\u91d1\u00b7\u7ef4\u683c\u7eb3\u57281960\u5e74\u66fe\u64b0\u6587\u201c\u6570\u5b66\u5728\u81ea\u7136\u79d1\u5b66\u4e2d\u4e0d\u5408\u7406\u7684\u6709\u6548\u6027\u201d<\/span>\uff08The Unreasonable Effectiveness of Mathematics in the Natural Sciences\uff09<\/span>\uff0c\u8868\u8fbe\u4ed6\u60ca\u5947\u4e8e\u6570\u5b66\u5bf9\u7269\u7406\u4e16\u754c\u7684\u6df1\u523b\u63cf\u8ff0\u80fd\u529b\uff0c\u6570\u5b66\u5e38\u5e38\u6307\u5f15\u7269\u7406\u7406\u8bba\u7684\u53d1\u5c55\uff0c\u751a\u81f3\u6307\u5f15\u5b9e\u9a8c\u9884\u6d4b\u3002\u8fd9\u9879\u65b0\u7814\u7a76\u7684\u8bba\u6587\u4f5c\u8005\u4e4b\u4e00 Amit Sharma \u5219\u53d1\u8868\u535a\u5ba2\u6587\u7ae0\u201c\u5927\u8bed\u8a00\u6a21\u578b\u5bf9\u4e8e\u56e0\u679c\u63a8\u7406\u4e0d\u5408\u7406\u7684\u6709\u6548\u6027\u201d\uff0c\u63a2\u8ba8\u5927\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u56e0\u679c\u63a8\u7406\u7684\u60ca\u4eba\u51c6\u786e\u6027\u3002<\/span><\/section>\n

<\/span><\/section>\n
\n
\u6587\u7ae0\u9898\u76ee\uff1aOn the unreasonable effectiveness of LLMs for causal inference<\/span><\/section>\n
\u6587\u7ae0\u94fe\u63a5\uff1ahttps:\/\/threadreaderapp.com\/thread\/1653457971844874240.html<\/span><\/section>\n<\/blockquote>\n

<\/section>\n
\u5728\u8fd9\u9879\u7814\u7a76\u4e2d\uff0c\u57fa\u4e8e\u5927\u8bed\u8a00\u6a21\u578b\u7684\u65b9\u6cd5\u5728\u591a\u4e2a\u56e0\u679c\u57fa\u51c6\u6d4b\u8bd5\u4efb\u52a1\u4e0a\u8868\u73b0\u51fa\u6700\u9ad8\u7684\u51c6\u786e\u6027<\/strong>\u3002\u57fa\u4e8eGPT-3.5\/4\u7684\u7b97\u6cd5\u5728\u591a\u9879\u56e0\u679c\u63a8\u7406\u4efb\u52a1\u4e2d\u80dc\u8fc7\u73b0\u6709\u7b97\u6cd5\uff0c\u5305\u62ec\u6210\u5bf9\u56e0\u679c\u53d1\u73b0\u4efb\u52a1<\/span>\uff0897%\uff0c\u63d0\u9ad813\u4e2a\u767e\u5206\u70b9\uff09<\/span>\uff0c\u53cd\u4e8b\u5b9e\u63a8\u7406\u4efb\u52a1<\/span>\uff0892%\uff0c\u63d0\u9ad820\u4e2a\u767e\u5206\u70b9\uff09<\/span>\uff0c\u548c\u5b9e\u9645\u56e0\u679c\u5173\u7cfb<\/span>\uff08\u5728\u786e\u5b9a\u4e8b\u4ef6\u7684\u5fc5\u8981\u548c\u5145\u5206\u539f\u56e0\u65b9\u9762\u5177\u670986%\u7684\u51c6\u786e\u6027\uff09<\/span>\u3002<\/span><\/section>\n

<\/span><\/section>\n
\u56fe\u7075\u5956\u5f97\u4e3b\u3001\u8ba1\u7b97\u673a\u79d1\u5b66\u5bb6 Judea Pearl \u5728\u63a8\u7279\u8f6c\u53d1\u8bba\u6587\u5e76\u8bc4\u8bba\u9053\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u5e94\u7528\u4e8e\u56e0\u679c\u63a8\u7406\u4efb\u52a1\u5177\u6709\u5de8\u5927\u7684\u6f5c\u529b\uff0c\u53ef\u80fd\u4e3a\u201c\u56e0\u679c\u4e4b\u68af\u201d\u6dfb\u52a0\u65b0\u7684\u5c42\u7ea7\u3002<\/span><\/section>\n

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

<\/p>\n

\u56fe1. Judea Pearl \u7684\u56e0\u679c\u4e4b\u68af\u5305\u62ec\u4e09\u4e2a\u5c42\u7ea7\uff1a\u5173\u8054\uff08association\uff09\u3001\u5e72\u9884\uff08intervention\uff09\u548c\u53cd\u4e8b\u5b9e\uff08counterfactual\uff09\uff0c\u5206\u522b\u5bf9\u5e94\u9010\u7ea7\u590d\u6742\u7684\u56e0\u679c\u95ee\u9898\u3002<\/span><\/p>\n


<\/span><\/section>\n
\u8fd9\u5230\u5e95\u662f\u5982\u4f55\u5b9e\u73b0\u7684\u5462\uff1f\u5173\u952e\u4e4b\u5904\u5728\u4e8e\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u5f15\u5165\u4e00\u79cd\u57fa\u4e8e\u6587\u672c\u548c\u5143\u6570\u636e\u7684\u65b0\u63a8\u7406\u65b9\u5f0f\u6765\u5b9e\u73b0\u8fd9\u4e00\u76ee\u6807\uff0c\u79f0\u4e4b\u4e3a\u57fa\u4e8e\u77e5\u8bc6\u7684\u56e0\u679c\u63a8\u7406<\/strong><\/span>\uff08knowledge-based causal reasoning\uff09<\/span>\uff0c\u8fd9\u4e0e\u73b0\u6709\u7684\u57fa\u4e8e\u6570\u636e\u7684\u65b9\u6cd5\u6709\u6240\u4e0d\u540c\u3002\u5177\u4f53\u800c\u8a00\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u62e5\u6709\u8fc4\u4eca\u4e3a\u6b62\u88ab\u8ba4\u4e3a\u53ea\u6709\u4eba\u7c7b\u624d\u5177\u6709\u7684\u80fd\u529b\uff0c\u5982\u4f7f\u7528\u77e5\u8bc6\u751f\u6210\u56e0\u679c\u56fe\uff0c\u6216\u4ece\u81ea\u7136\u8bed\u8a00\u4e2d\u8bc6\u522b\u80cc\u666f\u56e0\u679c\u5173\u7cfb\u3002<\/span><\/section>\n

<\/span><\/section>\n
\u5927\u8bed\u8a00\u6a21\u578b\u53ef\u4ee5\u4f5c\u4e3a\u4eba\u7c7b\u9886\u57df\u77e5\u8bc6\u7684\u4ee3\u7406\uff0c\u8fd9\u5bf9\u901a\u5e38\u4f9d\u8d56\u4e8e\u4eba\u7c7b\u8f93\u5165\u7684\u56e0\u679c\u4efb\u52a1\u6765\u8bf4\u662f\u4e00\u4e2a\u5de8\u5927\u7684\u80dc\u5229\u3002\u901a\u8fc7\u6355\u6349\u5173\u4e8e\u56e0\u679c\u673a\u5236\u7684\u5e38\u8bc6\u548c\u9886\u57df\u77e5\u8bc6\uff0c\u5e76\u652f\u6301\u81ea\u7136\u8bed\u8a00\u4e0e\u5f62\u5f0f\u65b9\u6cd5\u4e4b\u95f4\u7684\u8f6c\u6362\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u4e3a\u63a8\u8fdb\u56e0\u679c\u5173\u7cfb\u7684\u7814\u7a76\u3001\u5b9e\u8df5\u548c\u91c7\u7528\u5f00\u8f9f\u4e86\u65b0\u524d\u6cbf\u3002<\/span><\/section>\n

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

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

\n
\n
\n
\n
\n

<\/section>\n

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

1. \u5927\u8bed\u8a00\u6a21\u578b\u4e0e\u56e0\u679c\u53d1\u73b0<\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

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

<\/section>\n