{"id":20602,"date":"2024-06-17T14:07:12","date_gmt":"2024-06-17T06:07:12","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=20602"},"modified":"2024-06-17T14:07:12","modified_gmt":"2024-06-17T06:07:12","slug":"acl-2024%e8%ae%ba%e6%96%87%e7%9b%96%e6%a3%ba%e5%ae%9a%e8%ae%ba%ef%bc%9a%e5%a4%a7%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8b%e2%89%a0%e4%b8%96%e7%95%8c%e6%a8%a1%e6%8b%9f%e5%99%a8%ef%bc%8cyann-lecun%ef%bc%9a","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2024\/06\/17\/20602\/","title":{"rendered":"ACL 2024\u8bba\u6587\u76d6\u68fa\u5b9a\u8bba\uff1a\u5927\u8bed\u8a00\u6a21\u578b\u2260\u4e16\u754c\u6a21\u62df\u5668\uff0cYann LeCun\uff1a\u592a\u5bf9\u4e86"},"content":{"rendered":"
\"\"<\/div>\n
\u5982\u679c GPT-4 \u5728\u6a21\u62df\u57fa\u4e8e\u5e38\u8bc6\u4efb\u52a1\u7684\u72b6\u6001\u53d8\u5316\u65f6\u51c6\u786e\u7387\u90fd\u53ea\u6709\u7ea6 60%\uff0c\u90a3\u4e48\u6211\u4eec\u8fd8\u8981\u8003\u8651\u5c06\u5927\u8bed\u8a00\u6a21\u578b\u4f5c\u4e3a\u4e16\u754c\u6a21\u62df\u5668\u6765\u4f7f\u7528\u5417\uff1f<\/div>\n
\n
\n

\u6700\u8fd1\u4e24\u5929\uff0c\u4e00\u7bc7\u5165\u9009 ACL 2024 \u7684\u8bba\u6587\u300aCan Language Models Serve as Text-Based World Simulators?\u300b\u5728\u793e\u4ea4\u5a92\u4f53 X \u4e0a\u5f15\u53d1\u4e86\u70ed\u8bae\uff0c\u5c31\u8fde\u56fe\u7075\u5956\u5f97\u4e3b Yann LeCun \u4e5f\u53c2\u4e0e\u4e86\u8fdb\u6765\u3002<\/p>\n

\u8fd9\u7bc7\u8bba\u6587\u63a2\u8ba8\u7684\u95ee\u9898\u662f\uff1a\u5f53\u524d\u8bed\u8a00\u6a21\u578b\u672c\u8eab\u662f\u5426\u53ef\u4ee5\u5145\u5f53\u4e16\u754c\u6a21\u62df\u5668\uff0c\u5e76\u6b63\u786e\u9884\u6d4b\u52a8\u4f5c\u5982\u4f55\u6539\u53d8\u4e0d\u540c\u7684\u4e16\u754c\u72b6\u6001\uff0c\u4ece\u800c\u907f\u514d\u5927\u91cf\u624b\u52a8\u7f16\u7801\u7684\u9700\u8981\u5462<\/strong>\uff1f<\/p>\n

\u9488\u5bf9\u8fd9\u4e00\u95ee\u9898\uff0c\u6765\u81ea\u4e9a\u5229\u6851\u90a3\u5927\u5b66\u3001\u7ebd\u7ea6\u5927\u5b66\u3001\u7ea6\u7ff0\u65af\u30fb\u970d\u666e\u91d1\u65af\u5927\u5b66\u3001\u5fae\u8f6f\u7814\u7a76\u9662\u3001 \u827e\u4f26\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u6240\u7b49\u673a\u6784\u7684\u7814\u7a76\u8005\u5728\u300c\u57fa\u4e8e\u6587\u672c\u7684\u6a21\u62df\u5668\u300d\u4e0a\u4e0b\u6587\u4e2d\u7ed9\u51fa\u4e86\u4ed6\u4eec\u7684\u7b54\u6848\u3002<\/p>\n

\u4ed6\u4eec\u8ba4\u4e3a\uff1a\u8bed\u8a00\u6a21\u578b\u5e76\u4e0d\u80fd\u4f5c\u4e3a\u4e16\u754c\u6a21\u62df\u5668\u4f7f\u7528<\/strong>\u3002\u6bd4\u5982\uff0cGPT-4 \u5728\u6a21\u62df\u57fa\u4e8e\u5e38\u8bc6\u4efb\u52a1\uff08\u5982\u70e7\u5f00\u6c34\uff09\u7684\u72b6\u6001\u53d8\u5316\u65f6\uff0c\u51c6\u786e\u7387\u4ec5\u4e3a\u7ea6 60%\u3002<\/p>\n

\"\u56fe\u7247\"x \u5730\u5740\uff1ahttps:\/\/x.com\/peterjansen_ai\/status\/1801687501557665841<\/p>\n

Yann LeCun \u5bf9\u8fd9\u7bc7\u8bba\u6587\u7684\u53d1\u73b0\u8868\u793a\u4e86\u8ba4\u540c\uff0c\u5e76\u8ba4\u4e3a\u300c\u6ca1\u6709\u4e16\u754c\u6a21\u578b\uff0c\u4e5f\u5c31\u6ca1\u6709\u89c4\u5212\u3002\u300d<\/p>\n

\"\u56fe\u7247\"X \u5730\u5740\uff1ahttps:\/\/x.com\/ylecun\/status\/1801978192950927511<\/p>\n

\u4e0d\u8fc7\u4e5f\u6709\u4eba\u8868\u8fbe\u4e86\u4e0d\u540c\u7684\u89c2\u70b9\uff1a\u5f53\u524d LLM\uff08\u6ca1\u6709\u8fdb\u884c\u9488\u5bf9\u6027\u4efb\u52a1\u8bad\u7ec3\uff09\u7684\u51c6\u786e\u7387\u53ef\u4ee5\u8fbe\u5230 60\uff05\uff0c\u8fd9\u4e0d\u5c31\u8bf4\u660e\u4e86\u5b83\u4eec\u81f3\u5c11\u662f\u300c\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u7684\u4e16\u754c\u6a21\u578b\u300d\u5417\uff1f\u5e76\u4e14\u4f1a\u968f\u7740 LLM \u7684\u8fed\u4ee3\u800c\u6301\u7eed\u6539\u8fdb\u3002LeCun \u53c8\u8868\u793a\uff0c\u4e16\u754c\u6a21\u578b\u4e0d\u4f1a\u662f LLM\u3002<\/p>\n

\"\u56fe\u7247\"<\/p>\n

\u56de\u5230\u8bba\u6587\u4e2d\uff0c\u7814\u7a76\u8005\u6784\u5efa\u5e76\u4f7f\u7528\u4e86\u4e00\u4e2a\u5168\u65b0\u7684\u57fa\u51c6\uff0c\u4ed6\u4eec\u79f0\u4e3a\u300cByteSized32-State-Prediction\u300d\uff0c\u5305\u542b\u4e86\u4e00\u4e2a\u6587\u672c\u6e38\u620f\u72b6\u6001\u8f6c\u6362\u548c\u968f\u9644\u6e38\u620f\u4efb\u52a1\u7ec4\u6210\u7684\u6570\u636e\u96c6\u3002\u4ed6\u4eec\u9996\u6b21\u4f7f\u7528\u8be5\u57fa\u51c6\u6765\u76f4\u63a5\u91cf\u5316\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u4f5c\u4e3a\u57fa\u4e8e\u6587\u672c\u7684\u4e16\u754c\u6a21\u62df\u5668\u7684\u6027\u80fd\u3002<\/p>\n

\u901a\u8fc7\u5728\u8fd9\u4e2a\u6570\u636e\u96c6\u4e0a\u6d4b\u8bd5 GPT-4\uff0c\u7814\u7a76\u8005\u53d1\u73b0\uff1a\u5c3d\u7ba1\u5b83\u7684\u6027\u80fd\u4ee4\u4eba\u5370\u8c61\u6df1\u523b\uff0c\u4f46\u5982\u679c\u6ca1\u6709\u8fdb\u4e00\u6b65\u7684\u521b\u65b0\uff0c\u5b83\u4ecd\u7136\u662f\u4e00\u4e2a\u4e0d\u53ef\u9760\u7684\u4e16\u754c\u6a21\u62df\u5668\u3002<\/p>\n

\u56e0\u6b64\uff0c\u7814\u7a76\u8005\u8ba4\u4e3a\uff0c\u4ed6\u4eec\u7684\u5de5\u4f5c\u65e2\u4e3a\u5f53\u524d LLM \u7684\u80fd\u529b\u548c\u5f31\u70b9\u63d0\u4f9b\u4e86\u65b0\u7684\u89c1\u89e3\uff0c\u4e5f\u4e3a\u8ddf\u8e2a\u65b0\u6a21\u578b\u51fa\u73b0\u65f6\u7684\u672a\u6765\u8fdb\u5c55\u63d0\u4f9b\u4e86\u4e00\u4e2a\u65b0\u7684\u57fa\u51c6\u3002<\/p>\n

\"\u56fe\u7247\"<\/p>\n

\u8bba\u6587\u5730\u5740\uff1ahttps:\/\/arxiv.org\/pdf\/2406.06485<\/p>\n

\u65b9\u6cd5\u6982\u89c8<\/h4>\n

\u7814\u7a76\u8005\u63a2\u7a76\u4e86 LLM \u5728\u57fa\u4e8e\u6587\u672c\u7684\u865a\u62df\u73af\u5883\u4e2d\u5145\u5f53\u4e16\u754c\u6a21\u62df\u5668\u7684\u80fd\u529b\uff0c\u5728\u8fd9\u79cd\u73af\u5883\u4e2d\uff0c\u667a\u80fd\u4f53\u63a5\u6536\u89c2\u5bdf\u7ed3\u679c\u5e76\u4ee5\u81ea\u7136\u8bed\u8a00\u63d0\u51fa\u64cd\u4f5c\u4ee5\u5b8c\u6210\u67d0\u4e9b\u76ee\u6807\u3002<\/p>\n

\u6bcf\u4e2a\u6587\u672c\u73af\u5883\u90fd\u53ef\u4ee5\u6b63\u5f0f\u8868\u793a\u4e3a\u5177\u6709 7 \u5143\u7ec4 (S,A,T,O,R,C,D) \u7684\u76ee\u6807\u6761\u4ef6\u90e8\u5206\u53ef\u89c2\u5bdf\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b (POMDP)\uff0cS \u8868\u793a\u72b6\u6001\u7a7a\u95f4\uff0cA \u8868\u793a\u52a8\u4f5c\u7a7a\u95f4\uff0cT : S\u00d7A\u2192S \u8868\u793a\u8f6c\u6362\u51fd\u6570\uff0cO \u8868\u793a\u89c2\u5bdf\u51fd\u6570\uff0cR : S\u00d7A\u2192R \u8868\u793a\u5956\u52b1\u51fd\u6570\uff0cC \u8868\u793a\u63cf\u8ff0\u76ee\u6807\u548c\u52a8\u4f5c\u8bed\u4e49\u7684\u81ea\u7136\u8bed\u8a00\u300c\u4e0a\u4e0b\u6587\u6d88\u606f\u300d\uff0cD : S\u00d7A\u2192{0,1} \u8868\u793a\u4e8c\u5143\u5b8c\u6210\u6307\u793a\u51fd\u6570\u3002<\/p>\n

\u5927\u6a21\u578b\u6a21\u62df\u5668\uff08LLM-Sim\uff09\u4efb\u52a1<\/strong><\/p>\n

\u7814\u7a76\u8005\u63d0\u51fa\u4e86\u4e00\u4e2a\u9884\u6d4b\u4efb\u52a1\uff0c\u79f0\u5b83\u4e3a\u00a0LLM as-a-Simulator (LLM-Sim)\uff0c\u7528\u6765\u5b9a\u91cf\u8bc4\u4f30\u8bed\u8a00\u6a21\u578b\u4f5c\u4e3a\u53ef\u9760\u6a21\u62df\u5668\u7684\u80fd\u529b<\/strong>\u3002<\/p>\n

LLM-Sim \u4efb\u52a1\u662f\u5c06\u4e00\u4e2a\u51fd\u6570 F : C\u00d7S\u00d7A\u2192S\u00d7R\u00d7{0,1} \u4f5c\u4e3a\u4e16\u754c\u6a21\u62df\u5668\u6765\u5b9e\u73b0\u3002\u5728\u5b9e\u8df5\u4e2d\uff0c\u5b8c\u6574\u72b6\u6001\u8f6c\u6362\u6a21\u62df\u5668 F \u5e94\u8be5\u8003\u8651\u4e24\u79cd\u7c7b\u578b\u7684\u72b6\u6001\u8f6c\u6362\uff1a\u52a8\u4f5c\u9a71\u52a8\u8f6c\u6362\u548c\u73af\u5883\u9a71\u52a8\u8f6c\u6362<\/strong>\u3002<\/p>\n

\u56fe 1 \u4e3a\u4f7f\u7528 LLM \u4f5c\u4e3a\u6587\u672c\u6e38\u620f\u6a21\u62df\u5668\u7684\u793a\u4f8b\uff1a\u6253\u5f00\u6c34\u69fd\u540e\uff0c\u6c34\u69fd\u4e2d\u7684\u676f\u5b50\u88ab\u6c34\u586b\u6ee1\u7684\u8fc7\u7a0b\u3002\u52a8\u4f5c\u9a71\u52a8\u8f6c\u6362\u662f\u91c7\u53d6\u6253\u5f00\u6c34\u69fd\u7684\u52a8\u4f5c\u540e\uff0c\u6c34\u69fd\u88ab\u6253\u5f00\uff08isOn=true\uff09\uff1b\u800c\u73af\u5883\u9a71\u52a8\u8f6c\u6362\u662f\u5728\u6c34\u69fd\u6253\u5f00\u65f6\uff0c\u6c34\u586b\u6ee1\u6c34\u69fd\u4e2d\u7684\u676f\u5b50\u3002<\/p>\n

\"\u56fe\u7247\"<\/p>\n

\u4e3a\u4e86\u66f4\u597d\u5730\u7406\u89e3 LLM \u5bf9\u6bcf\u4e2a\u8f6c\u6362\u8fdb\u884c\u5efa\u6a21\u7684\u80fd\u529b\uff0c\u7814\u7a76\u8005\u8fdb\u4e00\u6b65\u5c06\u6a21\u62df\u5668\u51fd\u6570 F \u5206\u89e3\u4e3a\u4e09\u4e2a\u6b65\u9aa4\uff1a<\/p>\n

\"\u56fe\u7247\"<\/p>\n

    \n
  • \u52a8\u4f5c\u9a71\u52a8\u8f6c\u6362\u6a21\u62df\u5668<\/strong>\uff1a\u7ed9\u5b9a c\u3001s_t \u548c a_t\uff0c F_act\uff1aC\u00d7S\u00d7A\u2192S \u9884\u6d4b s^act_t+1\uff0c\u5176\u4e2d s^act_t+1 \u8868\u793a\u52a8\u4f5c\u5f15\u8d77\u7684\u76f4\u63a5\u72b6\u6001\u53d8\u5316\u3002<\/li>\n
  • \u73af\u5883\u9a71\u52a8\u8f6c\u6362\u6a21\u62df\u5668<\/strong>\uff1a\u7ed9\u5b9a c \u548c s^act_t+1\uff0cF_env\uff1aC\u00d7S\u2192S \u9884\u6d4b s_t+1\uff0c\u5176\u4e2d s_t+1 \u662f\u4efb\u4f55\u73af\u5883\u9a71\u52a8\u8f6c\u6362\u540e\u4ea7\u751f\u7684\u72b6\u6001\u3002<\/li>\n
  • \u6e38\u620f\u8fdb\u5ea6\u6a21\u62df\u5668<\/strong>\uff1a\u7ed9\u5b9a c\u3001s_t+1 \u548c a_t\uff0c F_R\uff1aC\u00d7S\u00d7A\u2192R\u00d7{0,1} \u9884\u6d4b\u5956\u52b1 r_t+1 \u548c\u6e38\u620f\u5b8c\u6210\u72b6\u6001 d_t+1\u3002<\/li>\n<\/ul>\n

    \u6b64\u5916\uff0c\u7814\u7a76\u8005\u8003\u8651\u4e86 LLM-Sim \u4efb\u52a1\u7684\u4e24\u79cd\u53d8\u4f53<\/p>\n

      \n
    • \u5b8c\u6574\u72b6\u6001\u9884\u6d4b<\/strong>\uff1aLLM \u8f93\u51fa\u5b8c\u6574\u72b6\u6001\u3002<\/li>\n
    • \u72b6\u6001\u5dee\u5f02\u9884\u6d4b<\/strong>\uff1aLLM \u4ec5\u8f93\u51fa\u8f93\u5165\u548c\u8f93\u51fa\u72b6\u6001\u4e4b\u95f4\u7684\u5dee\u5f02\u3002<\/li>\n<\/ul>\n

      \u6570\u636e\u548c\u8bc4\u4f30<\/strong><\/p>\n

      \u4e3a\u4e86\u5b8c\u6210\u8fd9\u4e00\u4efb\u52a1\uff0c\u7814\u7a76\u8005\u5f15\u5165\u4e86\u4e00\u4e2a\u65b0\u7684\u6587\u672c\u6e38\u620f\u72b6\u6001\u8f6c\u6362\u6570\u636e\u96c6\u3002\u8be5\u6570\u636e\u96c6\u4e3a \u300cBYTESIZED32-State-Prediction (BYTESIZED32-SP) \u300d\uff0c\u5b83\u5305\u542b 76,369 \u4e2a\u8f6c\u6362\uff0c\u8868\u793a\u4e3a (c,s_t,rt,d_t,a_t,s^act_t+1,s_t+1,r_t+1,d_t+1) \u5143\u7ec4<\/strong>\u3002\u8fd9\u4e9b\u8f6c\u6362\u662f\u4ece 31 \u4e2a\u4e0d\u540c\u7684\u6587\u672c\u6e38\u620f\u4e2d\u6536\u96c6\u7684\u3002<\/p>\n

      \u4e0b\u8868 1 \u603b\u7ed3\u4e86\u989d\u5916\u8bed\u6599\u5e93\u7edf\u8ba1\u6570\u636e\u3002<\/p>\n

      \"\u56fe\u7247\"<\/p>\n

      LLM-Sim \u4e0a\u7684\u6027\u80fd\u7531\u6a21\u578b\u76f8\u5bf9\u4e8e\u6d4b\u8bd5\u6837\u672c\u6570\u636e\u96c6\u4e0a\u7684\u771f\u5b9e\u6807\u7b7e\u7684\u9884\u6d4b\u51c6\u786e\u6027\u6765\u51b3\u5b9a\u3002\u6839\u636e\u5b9e\u9a8c\u6761\u4ef6\uff0cLLM \u5fc5\u987b\u6a21\u62df\u5bf9\u8c61\u5c5e\u6027\uff08\u6a21\u62df F_act\u3001F_env \u6216 F\uff09\u548c \/ \u6216\u6e38\u620f\u8fdb\u5ea6\uff08\u6a21\u62df F_R \u6216 F\uff09\uff0c\u5b9a\u4e49\u5982\u4e0b\uff1a<\/p>\n

        \n
      • \u5bf9\u8c61\u5c5e\u6027<\/strong>\uff1a\u6e38\u620f\u4e2d\u6240\u6709\u5bf9\u8c61\u3001\u6bcf\u4e2a\u5bf9\u8c61\u7684\u5c5e\u6027\uff08\u5982\u6e29\u5ea6\u3001\u5927\u5c0f\uff09\uff0c\u4ee5\u53ca\u4e0e\u5176\u4ed6\u5bf9\u8c61\u7684\u5173\u7cfb\uff08\u5982\u5728\u53e6\u4e00\u4e2a\u5bf9\u8c61\u5185\u6216\u4e4b\u4e0a\uff09\u3002<\/li>\n
      • \u6e38\u620f\u8fdb\u5ea6<\/strong>\uff1a\u667a\u80fd\u4f53\u76f8\u5bf9\u4e8e\u603b\u4f53\u76ee\u6807\u7684\u72b6\u6001\uff0c\u5305\u62ec\u5f53\u524d\u7d2f\u79ef\u7684\u5956\u52b1\u3001\u6e38\u620f\u662f\u5426\u5df2\u7ec8\u6b62\u4ee5\u53ca\u603b\u4f53\u76ee\u6807\u662f\u5426\u5df2\u5b9e\u73b0\u3002<\/li>\n<\/ul>\n

        \u7814\u7a76\u8005\u6ce8\u610f\u5230\uff0c\u5728\u6bcf\u79cd\u60c5\u51b5\u4e0b\uff0cLLM \u90fd\u63d0\u4f9b\u4e86 ground truth \u5148\u524d\u72b6\u6001\uff08\u5f53\u51fd\u6570\u4e3a F_env \u65f6\uff0c\u5148\u524d\u72b6\u6001\u4e3a s^act_t+1 \uff09\u4ee5\u53ca\u6574\u4f53\u4efb\u52a1\u4e0a\u4e0b\u6587\u3002\u4e5f\u5c31\u662f\u8bf4\uff0cLLM \u59cb\u7ec8\u6267\u884c\u5355\u6b65\u9884\u6d4b\u3002<\/p>\n

        \u5b9e\u9a8c\u7ed3\u679c<\/h4>\n

        \u4e0a\u56fe 1 \u6f14\u793a\u4e86\u7814\u7a76\u8005\u4f7f\u7528\u4e0a\u4e0b\u6587\u5b66\u4e60\u8bc4\u4f30 LLM-Sim \u4efb\u52a1\u4e2d\u6a21\u578b\u7684\u6027\u80fd\u3002\u4ed6\u4eec\u8bc4\u4f30\u4e86 GPT-4 \u5728\u5b8c\u6574\u72b6\u6001\u548c\u72b6\u6001\u5dee\u5f02\u9884\u6d4b\u673a\u5236\u4e2d\u7684\u51c6\u786e\u6027\u3002\u8be5\u6a21\u578b\u63a5\u6536\u5148\u524d\u72b6\u6001\uff08\u7f16\u7801\u4e3a JSON \u5bf9\u8c61\uff09\u3001\u5148\u524d\u64cd\u4f5c\u548c\u4e0a\u4e0b\u6587\u6d88\u606f\uff0c\u5e76\u751f\u6210\u540e\u7eed\u72b6\u6001\uff08\u4f5c\u4e3a\u5b8c\u6574\u7684 JSON \u5bf9\u8c61\u6216\u5dee\u5f02\uff09\u3002<\/p>\n

        \u4e0b\u8868 2 \u5c55\u793a\u4e86 GPT-4 \u6a21\u62df\u5b8c\u6574\u72b6\u6001\u8f6c\u6362\u7684\u51c6\u786e\u6027\uff0c\u4ee5\u53ca\u5355\u72ec\u6a21\u62df\u52a8\u4f5c\u9a71\u52a8\u8f6c\u6362\u548c\u73af\u5883\u9a71\u52a8\u8f6c\u6362\u7684\u51c6\u786e\u6027\u3002<\/p>\n

        \"\u56fe\u7247\"<\/p>\n

        \u7814\u7a76\u8005\u5f97\u51fa\u4e86\u4ee5\u4e0b\u51e0\u9879\u91cd\u8981\u53d1\u73b0\uff1a<\/p>\n

        \u9884\u6d4b\u52a8\u4f5c\u9a71\u52a8\u8f6c\u6362\u6bd4\u9884\u6d4b\u73af\u5883\u9a71\u52a8\u8f6c\u6362\u66f4\u5bb9\u6613<\/strong>\u3002\u5728\u6700\u597d\u7684\u60c5\u51b5\u4e0b\uff0cGPT-4 \u80fd\u591f\u6b63\u786e\u6a21\u62df 77.1% \u7684\u52a8\u6001\u52a8\u4f5c\u9a71\u52a8\u8f6c\u6362\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cGPT-4 \u6700\u591a\u53ea\u80fd\u6b63\u786e\u6a21\u62df 49.7% \u7684\u52a8\u6001\u73af\u5883\u9a71\u52a8\u8f6c\u6362\u3002<\/p>\n

        \u9884\u6d4b\u9759\u6001\u8f6c\u6362\u6bd4\u52a8\u6001\u8f6c\u6362\u66f4\u5bb9\u6613<\/strong>\u3002\u4e0d\u51fa\u6240\u6599\uff0c\u5728\u5927\u591a\u6570\u60c5\u51b5\u4e0b\uff0c\u5bf9\u9759\u6001\u8f6c\u6362\u8fdb\u884c\u5efa\u6a21\u6bd4\u5bf9\u52a8\u6001\u8f6c\u6362\u8fdb\u884c\u5efa\u6a21\u8981\u5bb9\u6613\u5f97\u591a\u3002<\/p>\n

        \u5bf9\u4e8e\u52a8\u6001\u72b6\u6001\uff0c\u9884\u6d4b\u5b8c\u6574\u6e38\u620f\u72b6\u6001\u66f4\u5bb9\u6613\uff1b\u800c\u5bf9\u4e8e\u9759\u6001\u72b6\u6001\uff0c\u9884\u6d4b\u72b6\u6001\u5dee\u5f02\u66f4\u5bb9\u6613\u3002\u9884\u6d4b\u52a8\u6001\u72b6\u6001\u7684\u72b6\u6001\u5dee\u5f02\u53ef\u4ee5\u663e\u8457\u63d0\u9ad8\u6a21\u62df\u9759\u6001\u8f6c\u6362\u7684\u6027\u80fd\uff08>10%\uff09\uff0c\u800c\u6a21\u62df\u52a8\u6001\u8f6c\u6362\u65f6\u7684\u6027\u80fd\u4f1a\u964d\u4f4e\u3002<\/p>\n

        \u6e38\u620f\u89c4\u5219\u5f88\u91cd\u8981\uff0cLLM \u80fd\u591f\u751f\u6210\u8db3\u591f\u597d\u7684\u6e38\u620f\u89c4\u5219<\/strong>\u3002\u5f53\u4e0a\u4e0b\u6587\u6d88\u606f\u4e2d\u672a\u63d0\u4f9b\u6e38\u620f\u89c4\u5219\u65f6\uff0cGPT-4 \u5728\u6240\u6709\u4e09\u4e2a\u6a21\u62df\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\u5728\u5927\u591a\u6570\u60c5\u51b5\u4e0b\u90fd\u4f1a\u4e0b\u964d\u3002<\/p>\n

        GPT-4 \u5728\u5927\u591a\u6570\u60c5\u51b5\u4e0b\u90fd\u80fd\u9884\u6d4b\u6e38\u620f\u8fdb\u5ea6<\/strong>\u3002\u4e0b\u8868 3 \u5c55\u793a\u4e86 GPT-4 \u9884\u6d4b\u6e38\u620f\u8fdb\u5ea6\u7684\u7ed3\u679c\u3002\u5728\u4e0a\u4e0b\u6587\u4e2d\u52a0\u5165\u4e86\u6e38\u620f\u89c4\u5219\u4fe1\u606f\u540e\uff0cGPT-4 \u53ef\u4ee5\u5728 92.1% \u7684\u6d4b\u8bd5\u7528\u4f8b\u4e2d\u6b63\u786e\u9884\u6d4b\u6e38\u620f\u8fdb\u5ea6\u3002\u8fd9\u4e9b\u89c4\u5219\u7684\u5b58\u5728\u5728\u4e0a\u4e0b\u6587\u4e2d\u81f3\u5173\u91cd\u8981\uff1a\u5982\u679c\u6ca1\u6709\u5b83\u4eec\uff0cGPT-4 \u7684\u9884\u6d4b\u51c6\u786e\u7387\u4f1a\u4e0b\u964d\u5230 61.5%\u3002<\/p>\n

        \"\u56fe\u7247\"<\/p>\n

        \u4eba\u7c7b\u5728 LLM-Sim \u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u4f18\u4e8e GPT-4<\/strong>\u3002\u7814\u7a76\u8005\u5bf9 LLM-Sim \u4efb\u52a1\u8fdb\u884c\u4e86\u521d\u6b65\u7684\u4eba\u7c7b\u7814\u7a76\u3002\u7ed3\u679c\u89c1\u4e0b\u8868 4\u3002<\/p>\n

        \u7ed3\u679c\u53d1\u73b0\uff0c\u4eba\u7c7b\u7684\u6574\u4f53\u51c6\u786e\u7387\u4e3a 80%\uff0c\u800c\u91c7\u6837\u7684 LLM \u7684\u51c6\u786e\u7387\u4e3a 50%\uff0c\u5e76\u4e14\u4e0d\u540c\u6ce8\u91ca\u8005\u4e4b\u95f4\u7684\u5dee\u5f02\u5f88\u5c0f\u3002\u8fd9\u8868\u660e\uff0c\u867d\u7136\u4efb\u52a1\u5bf9\u4e8e\u4eba\u7c7b\u6765\u8bf4\u603b\u4f53\u4e0a\u662f\u76f4\u89c2\u4e14\u76f8\u5bf9\u5bb9\u6613\u7684\uff0c\u4f46\u5bf9\u4e8e LLM \u6765\u8bf4\u4ecd\u6709\u5f88\u5927\u7684\u6539\u8fdb\u7a7a\u95f4\u3002<\/p>\n

        \"\u56fe\u7247\"<\/p>\n

        GPT-4 \u5728\u9700\u8981\u7b97\u672f\u3001\u5e38\u8bc6\u6216\u79d1\u5b66\u77e5\u8bc6\u65f6\u66f4\u5bb9\u6613\u51fa\u9519<\/strong>\u3002\u4e0b\u56fe 2 \u5c55\u793a\u4e86\u5728\u6574\u4f53\u72b6\u6001\u8f6c\u6362\u3001\u52a8\u4f5c\u9a71\u52a8\u8f6c\u6362\u548c\u73af\u5883\u9a71\u52a8\u8f6c\u6362\u4e2d\uff0c\u9884\u6d4b\u7ed3\u679c\u7684\u6b63\u786e\u6bd4\u4f8b\u3001\u5c06\u5c5e\u6027\u8bbe\u7f6e\u4e3a\u4e0d\u6b63\u786e\u503c\u7684\u6bd4\u4f8b\u6216\u672a\u80fd\u66f4\u6539\u5c5e\u6027\u503c\u7684\u6bd4\u4f8b\u3002<\/p>\n

        \u6211\u4eec\u53ef\u4ee5\u89c2\u5bdf\u5230\uff0cGPT-4 \u80fd\u591f\u5f88\u597d\u5730\u5904\u7406\u5927\u591a\u6570\u7b80\u5355\u7684\u5e03\u5c14\u503c\u5c5e\u6027\u3002\u9519\u8bef\u96c6\u4e2d\u51fa\u73b0\u5728\u9700\u8981\u7b97\u672f\uff08\u5982\u6e29\u5ea6\u3001timeAboveMaxTemp\uff09\u3001\u5e38\u8bc6\uff08\u5982 current_aperture\u3001current_focus\uff09\u6216\u79d1\u5b66\u77e5\u8bc6\uff08\u5982 on\uff09\u7684\u975e\u5e73\u51e1\u5c5e\u6027\u4e0a\u3002<\/p>\n

        \"\u56fe\u7247\"<\/p>\n

        \u66f4\u591a\u6280\u672f\u7ec6\u8282\u548c\u5b9e\u9a8c\u7ed3\u679c\u8bf7\u53c2\u9605\u539f\u8bba\u6587\u3002<\/p>\n<\/div>\n<\/div>\n

        <\/div>\n

        <\/g><\/svg><\/i> \"Loading\"<\/p>\n

        <\/div>\n","protected":false},"excerpt":{"rendered":"

        \u5982\u679c GPT-4 \u5728\u6a21\u62df\u57fa\u4e8e\u5e38\u8bc6\u4efb\u52a1\u7684\u72b6\u6001\u53d8\u5316\u65f6\u51c6\u786e\u7387\u90fd\u53ea\u6709\u7ea6 60%\uff0c\u90a3\u4e48\u6211\u4eec\u8fd8\u8981\u8003\u8651\u5c06\u5927\u8bed\u8a00\u6a21\u578b\u4f5c\u4e3a\u4e16\u754c\u6a21 […]<\/p>\n

        <\/div>\n

        <\/g><\/svg><\/i> \"Loading\"<\/p>\n

        <\/div>\n","protected":false},"author":56,"featured_media":20604,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[23,80],"tags":[1150],"class_list":["post-20602","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-23","category-80","tag-gpt-4-"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/20602","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=20602"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/20602\/revisions"}],"predecessor-version":[{"id":20605,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/20602\/revisions\/20605"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/20604"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=20602"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=20602"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=20602"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}