{"id":35387,"date":"2024-12-24T10:59:55","date_gmt":"2024-12-24T02:59:55","guid":{"rendered":"https:\/\/17aitech.com\/?p=35387"},"modified":"2024-12-24T10:59:55","modified_gmt":"2024-12-24T02:59:55","slug":"%e8%bf%88%e5%90%91%e3%80%8c%e5%a4%9a%e9%9d%a2%e6%89%8b%e3%80%8d%e5%8c%bb%e7%96%97%e5%a4%a7%e6%a8%a1%e5%9e%8b%ef%bc%8c%e4%b8%8a%e4%ba%a4%e5%a4%a7%e5%9b%a2%e9%98%9f%e5%8f%91%e5%b8%83%e5%a4%a7%e8%a7%84","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=35387","title":{"rendered":"\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21\u578b\uff0c\u4e0a\u4ea4\u5927\u56e2\u961f\u53d1\u5e03\u5927\u89c4\u6a21\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u5168\u9762\u57fa\u51c6\u6d4b\u8bd5"},"content":{"rendered":"

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21\u578b\uff0c\u4e0a\u4ea4\u5927\u56e2\u961f\u53d1\u5e03\u5927\u89c4\u6a21\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u5168\u9762\u57fa\u51c6\u6d4b\u8bd5<\/a><\/p>\n

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

\u7f16\u8f91 |\u00a0ScienceAI<\/p>\n

\u8fd1\u65e5\uff0c\u4e0a\u6d77\u4ea4\u901a\u5927\u5b66\u3001\u4e0a\u6d77 AI Lab\u3001\u4e2d\u56fd\u79fb\u52a8\u7b49\u673a\u6784\u7684\u8054\u5408\u7814\u7a76\u56e2\u961f\uff0c\u5728 arXiv \u9884\u5370\u5e73\u53f0\u53d1\u5e03\u6587\u7ae0\u300aTowards Evaluating and Building Versatile Large Language Models for Medicine<\/em>\u300b\uff0c\u4ece\u6570\u636e\u3001\u6d4b\u8bc4\u3001\u6a21\u578b\u591a\u4e2a\u89d2\u5ea6\u5168\u9762\u5206\u6790\u8ba8\u8bba\u4e86\u4e34\u5e8a\u533b\u5b66\u5927\u8bed\u8a00\u6a21\u578b\u5e94\u7528\u3002<\/p>\n

\u6587\u4e2d\u6240\u6d89\u53ca\u7684\u6240\u6709\u6570\u636e\u548c\u4ee3\u7801\u3001\u6a21\u578b\u5747\u5df2\u5f00\u6e90\u3002<\/p>\n

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

GitHub: \u00a0https:\/\/github.com\/MAGIC-AI4Med\/MedS-Ins<\/em><\/section>\n
Paper Link: \u00a0https:\/\/arxiv.org\/abs\/2408.12547<\/em><\/section>\n
Leaderboard: \u00a0https:\/\/henrychur.github.io\/MedS-Bench\/<\/em><\/section>\n

\u6982\u89c8<\/strong><\/p>\n

\u8fd1\u5e74\u6765\uff0c\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u53d6\u5f97\u4e86\u663e\u8457\u7684\u8fdb\u5c55\uff0c\u5e76\u5728\u533b\u7597\u9886\u57df\u53d6\u5f97\u4e86\u4e00\u5b9a\u6210\u679c\u3002\u8fd9\u4e9b\u6a21\u578b\u5728\u533b\u5b66\u591a\u9879\u9009\u62e9\u95ee\u7b54\uff08MCQA\uff09\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u5c55\u73b0\u51fa\u9ad8\u6548\u7684\u80fd\u529b\uff0c\u5e76\u4e14 UMLS \u7b49\u4e13\u4e1a\u8003\u8bd5\u4e2d\u8fbe\u5230\u6216\u8d85\u8fc7\u4e13\u5bb6\u6c34\u5e73\u3002<\/p>\n

\u7136\u800c\uff0cLLM \u8ddd\u79bb\u5b9e\u9645\u4e34\u5e8a\u573a\u666f\u4e2d\u7684\u5e94\u7528\u4ecd\u7136\u6709\u76f8\u5f53\u957f\u7684\u8ddd\u79bb\u3002\u5176\u4e3b\u8981\u95ee\u9898\uff0c\u96c6\u4e2d\u5728\u6a21\u578b\u5728\u5904\u7406\u57fa\u672c\u533b\u5b66\u77e5\u8bc6\u65b9\u9762\u7684\u4e0d\u8db3\uff0c\u5982\u5728\u89e3\u8bfb\u00a0ICD<\/span> \u7f16\u7801\u3001\u9884\u6d4b\u4e34\u5e8a\u7a0b\u5e8f\u4ee5\u53ca\u89e3\u6790\u7535\u5b50\u5065\u5eb7\u8bb0\u5f55\uff08EHR\uff09\u6570\u636e\u65b9\u9762\u7684\u8bef\u5dee\u3002<\/p>\n

\u8fd9\u4e9b\u95ee\u9898\u6307\u5411\u4e86\u4e00\u4e2a\u5173\u952e\uff1a\u5f53\u524d\u7684\u8bc4\u4f30\u57fa\u51c6\u4e3b\u8981\u5173\u6ce8\u4e8e\u533b\u5b66\u8003\u8bd5\u9009\u62e9\u9898\uff0c\u800c\u4e0d\u80fd\u5145\u5206\u53cd\u6620 LLM \u5728\u771f\u5b9e\u4e34\u5e8a\u60c5\u666f\u4e2d\u7684\u5e94\u7528\u3002<\/p>\n

\u672c\u7814\u7a76\u63d0\u51fa\u4e86\u4e00\u9879\u65b0\u7684\u8bc4\u4f30\u57fa\u51c6 MedS-Bench\uff0c\u8be5\u57fa\u51c6\u4e0d\u4ec5\u5305\u62ec\u591a\u9879\u9009\u62e9\u9898\uff0c\u8fd8\u6db5\u76d6\u4e86\u4e34\u5e8a\u62a5\u544a\u6458\u8981\u3001\u6cbb\u7597\u5efa\u8bae\u3001\u8bca\u65ad\u548c\u547d\u540d\u5b9e\u4f53\u8bc6\u522b\u7b49 11 \u9879\u9ad8\u7ea7\u4e34\u5e8a\u4efb\u52a1\u3002<\/p>\n

\u7814\u7a76\u56e2\u961f\u901a\u8fc7\u6b64\u57fa\u51c6\u5bf9\u591a\u4e2a\u4e3b\u6d41\u7684\u533b\u7597\u6a21\u578b\u8fdb\u884c\u4e86\u8bc4\u4f30\uff0c\u53d1\u73b0\u5373\u4fbf\u662f\u4f7f\u7528\u4e86 few-shot prompting\uff0c\u6700\u5148\u8fdb\u6a21\u578b\uff0c\u4f8b\u5982\uff0cGPT-4\uff0cClaude<\/span> \u7b49\uff0c\u5728\u5904\u7406\u8fd9\u4e9b\u590d\u6742\u7684\u4e34\u5e8a\u4efb\u52a1\u65f6\u4e5f\u9762\u4e34\u56f0\u96be\u3002<\/p>\n

\u4e3a\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0c\u53d7\u5230 Super-NaturalInstructions \u7684\u542f\u53d1\uff0c\u7814\u7a76\u56e2\u961f\u6784\u5efa\u4e86\u9996\u4e2a\u5168\u9762\u7684\u533b\u5b66\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u96c6 MedS-Ins\uff0c\u8be5\u6570\u636e\u96c6\u6574\u5408\u4e86\u6765\u81ea\u8003\u8bd5\u3001\u4e34\u5e8a\u6587\u672c\u3001\u5b66\u672f\u8bba\u6587\u3001\u533b\u5b66\u77e5\u8bc6\u5e93\u53ca\u65e5\u5e38\u5bf9\u8bdd\u7684 58 \u4e2a\u751f\u7269\u533b\u5b66\u6587\u672c\u6570\u636e\u96c6\uff0c\u5305\u542b\u8d85\u8fc7 1350 \u4e07\u4e2a\u6837\u672c\uff0c\u6db5\u76d6\u4e86 122 \u4e2a\u4e34\u5e8a\u4efb\u52a1\u3002<\/p>\n

\u5728\u6b64\u57fa\u7840\u4e0a\uff0c\u7814\u7a76\u56e2\u961f\u5bf9\u5f00\u6e90\u533b\u5b66\u8bed\u8a00\u6a21\u578b\u8fdb\u884c\u6307\u4ee4\u8c03\u6574\uff0c\u63a2\u7d22\u4e86 in-context learning \u73af\u5883\u4e0b\u7684\u6a21\u578b\u6548\u679c\u3002<\/p>\n

\u8be5\u5de5\u4f5c\u4e2d\u5f00\u53d1\u7684\u533b\u5b66\u5927\u8bed\u8a00\u6a21\u578b\u2014\u2014MMedIns-Llama 3\uff0c\u5728\u591a\u79cd\u4e34\u5e8a\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u8d85\u8fc7\u4e86\u73b0\u6709\u7684\u9886\u5148\u95ed\u6e90\u6a21\u578b\uff0c\u5982 GPT-4 \u548c Claude-3.5\u3002MedS-Ins \u7684\u6784\u5efa\u6781\u5927\u7684\u4fc3\u8fdb\u4e86\u533b\u5b66\u5927\u8bed\u8a00\u6a21\u578b\u5728\u5b9e\u9645\u4e34\u5e8a\u573a\u666f\u7684\u4e2d\u7684\u80fd\u529b\uff0c\u4f7f\u5176\u5e94\u7528\u8303\u56f4\u8fdc\u8d85\u5728\u7ebf\u804a\u5929\u6216\u591a\u9879\u9009\u62e9\u95ee\u7b54\u7684\u9650\u5236\u3002<\/p>\n

\u76f8\u4fe1\u8fd9\u4e00\u8fdb\u5c55\u4e0d\u4ec5\u63a8\u52a8\u4e86\u533b\u5b66\u8bed\u8a00\u6a21\u578b\u7684\u53d1\u5c55\uff0c\u4e5f\u4e3a\u672a\u6765\u4e34\u5e8a\u5b9e\u8df5\u4e2d\u7684\u4eba\u5de5\u667a\u80fd\u5e94\u7528\u63d0\u4f9b\u4e86\u65b0\u7684\u53ef\u80fd\u6027\u3002<\/p>\n

\u6d4b\u8bd5\u57fa\u51c6\u6570\u636e\u96c6\uff08<\/strong>MedS-Bench\uff09<\/strong><\/p>\n

\u4e3a\u4e86\u8bc4\u4f30\u5404\u79cd LLM \u5728\u4e34\u5e8a\u5e94\u7528\u4e2d\u7684\u80fd\u529b\uff0c\u7814\u7a76\u56e2\u961f\u5f00\u53d1\u4e86 MedS-Bench\uff0c\u8fd9\u662f\u4e00\u4e2a\u8d85\u8d8a\u4f20\u7edf\u9009\u62e9\u9898\u7684\u7efc\u5408\u6027\u533b\u5b66\u57fa\u51c6\u3002\u5982\u4e0b\u56fe\u6240\u793a\uff0cMedS-Bench \u6e90\u81ea 39 \u4e2a\u73b0\u6709\u6570\u636e\u96c6\uff0c\u8986\u76d6 11 \u4e2a\u7c7b\u522b\uff0c\u603b\u5171\u5305\u542b 52 \u4e2a\u4efb\u52a1\u3002<\/p>\n

\u5728 MedS-Bench \u4e2d\uff0c\u6570\u636e\u88ab\u91cd\u65b0\u683c\u5f0f\u5316\u4e3a\u6307\u4ee4\u5fae\u8c03\u7684\u7ed3\u6784\u3002\u6b64\u5916\uff0c\u6bcf\u6761\u4efb\u52a1\u90fd\u914d\u6709\u4eba\u5de5\u6807\u6ce8\u7684\u4efb\u52a1\u5b9a\u4e49\u3002\u6d89\u53ca\u7684 11 \u4e2a\u7c7b\u522b\u5206\u522b\u662f\uff1a\u9009\u62e9\u9898\u89e3\u7b54\uff08MCQA\uff09\u3001\u6587\u672c\u6458\u8981\uff08Text Summarization\uff09\u3001\u4fe1\u606f\u63d0\u53d6\uff08Information Extraction<\/span>\uff09\u3001\u89e3\u91ca\u4e0e\u63a8\u7406\uff08Explanation and Rationale\uff09\u3001\u547d\u540d\u5b9e\u4f53\u8bc6\u522b\uff08NER\uff09\u3001\u8bca\u65ad\uff08Diagnosis\uff09\u3001\u6cbb\u7597\u8ba1\u5212\u89c4\u5212\uff08Treatment Planning\uff09\u3001\u4e34\u5e8a\u7ed3\u679c\u9884\u6d4b\uff08Clinical Outcome Prediction\uff09\u3001\u6587\u672c\u5206\u7c7b\uff08Text Classification\uff09\u3001\u4e8b\u5b9e\u9a8c\u8bc1\uff08Fact Verification\uff09\u548c\u81ea\u7136\u8bed\u8a00\u63a8\u7406\uff08NLI\uff09\u3002<\/p>\n

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

\u56fe 1\uff1aMedS-Bench \u6570\u636e\u7edf\u8ba1\u3002\u6b64\u5c42\u6b21\u73af\u5f62\u56fe\u8be6\u7ec6\u5730\u5c55\u793a\u4e86\u8bc4\u4f30\u57fa\u51c6\u4e2d\u7684\u6570\u636e\u5206\u5e03\u3002\u7b2c\u4e00\u5c42\u5206\u7c7b\u4e86\u4efb\u52a1\u7c7b\u578b\uff0c\u5305\u542b\u4e86 11 \u4e2a\u4e3b\u8981\u4efb\u52a1\u7c7b\u522b\u7684\u57fa\u51c6\u3002\u7b2c\u4e8c\u5c42\u8be6\u7ec6\u5217\u51fa\u4e86\u6d89\u53ca\u7684\u6570\u636e\u96c6\uff0c\u603b\u5171\u6709 39 \u4e2a\u6570\u636e\u96c6\u3002\u7b2c\u4e09\u5c42\u8be6\u7ec6\u8bf4\u660e\u4e86\u5177\u4f53\u7684\u4efb\u52a1\uff0c\u57fa\u51c6\u6d4b\u8bd5\u5171\u6db5\u76d6\u4e86 52 \u4e2a\u4e0d\u540c\u7684\u4efb\u52a1\u3002\u603b\u4f53\u800c\u8a00\uff0c\u8fd9\u4e00\u57fa\u51c6\u6d4b\u8bd5\u80fd\u591f\u5bf9\u6a21\u578b\u5728\u591a\u4e2a\u7ef4\u5ea6\u4e0a\u7684\u8868\u73b0\u8fdb\u884c\u5168\u9762\u800c\u6df1\u5165\u7684\u8bc4\u4f30\u3002<\/section>\n

\u9664\u4e86\u5b9a\u4e49\u8fd9\u4e9b\u4efb\u52a1\u7c7b\u522b\uff0c\u7814\u7a76\u56e2\u961f\u8fd8\u5bf9 MedS-Bench \u6587\u672c\u957f\u5ea6\u8fdb\u884c\u4e86\u8be6\u7ec6\u7684\u7edf\u8ba1\uff0c\u5e76\u533a\u5206\u4e86 LLM \u5904\u7406\u4e0d\u540c\u4efb\u52a1\u6240\u9700\u7684\u80fd\u529b\uff0c\u5982\u4e0b\u8868\u6240\u793a\u3002LLM \u5904\u7406\u4efb\u52a1\u6240\u9700\u7684\u80fd\u529b\u88ab\u5206\u4e3a\u4e24\u7c7b\uff1a(i)\u6839\u636e\u6a21\u578b\u5185\u90e8\u77e5\u8bc6\u8fdb\u884c\u63a8\u7406\uff1b(ii) \u4ece\u63d0\u4f9b\u7684\u4e0a\u4e0b\u6587\u4e2d\u68c0\u7d22\u4e8b\u5b9e\u3002<\/p>\n

\u5e7f\u4e49\u4e0a\u8bb2\uff0c\u524d\u8005\u6d89\u53ca\u7684\u4efb\u52a1\u9700\u8981\u4ece\u5927\u89c4\u6a21\u9884\u8bad\u7ec3\u4e2d\u83b7\u53d6\u7f16\u7801\u5728\u6a21\u578b\u6743\u91cd\u4e2d\u7684\u77e5\u8bc6\uff0c\u800c\u540e\u8005\u6d89\u53ca\u7684\u4efb\u52a1\u5219\u9700\u8981\u4ece\u6240\u63d0\u4f9b\u7684\u4e0a\u4e0b\u6587\u4e2d\u63d0\u53d6\u4fe1\u606f\uff0c\u5982\u603b\u7ed3\u6216\u4fe1\u606f\u63d0\u53d6\u3002\u5982\u8868 1 \u6240\u793a\uff0c\u603b\u5171\u6709\u516b\u7c7b\u4efb\u52a1\u8981\u6c42\u6a21\u578b\u4ece\u6a21\u578b\u4e2d\u8c03\u7528\u77e5\u8bc6\uff0c\u800c\u5176\u4f59\u4e09\u7c7b\u4efb\u52a1\u5219\u8981\u6c42\u4ece\u7ed9\u5b9a\u4e0a\u4e0b\u6587\u4e2d\u68c0\u7d22\u4e8b\u5b9e\u3002<\/p>\n

\u8868 1\uff1a\u6240\u7528\u6d4b\u8bd5\u4efb\u52a1\u7684\u8be6\u7ec6\u7edf\u8ba1\u4fe1\u606f\u3002<\/p>\n

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

\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u96c6\uff08<\/strong>MedS-Ins\uff09<\/strong><\/p>\n

\u6b64\u5916\uff0c\u7814\u7a76\u56e2\u961f\u8fd8\u5f00\u6e90\u4e86\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u96c6 MedS-Ins\u3002\u8be5\u6570\u636e\u96c6\u8986\u76d6 5 \u4e2a\u4e0d\u540c\u7684\u6587\u672c\u6e90\u548c 19 \u4e2a\u4efb\u52a1\u7c7b\u522b\uff0c\u5171\u8ba1 122 \u4e2a\u4e0d\u540c\u7684\u4e34\u5e8a\u4efb\u52a1\u3002\u4e0b\u56fe\u603b\u7ed3\u4e86 MedS-Ins\u7684\u6784\u9020\u6d41\u7a0b\u4ee5\u53ca\u7edf\u8ba1\u4fe1\u606f\u3002<\/p>\n

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

\u56fe 2\uff1aMedS-Ins \u6982\u89c8\u3002a \u4efb\u52a1\u6536\u96c6\u6d41\u7a0b\u3002\u5bf9\u4e8e\u6bcf\u4e2a\u4efb\u52a1\uff0c\u4eba\u5de5\u6807\u6ce8\u4e00\u4e2a\u4efb\u52a1\u7c7b\u522b\u548c\u4efb\u52a1\u5b9a\u4e49\uff0c\u5171\u8ba119\u4e2a\u4efb\u52a1\u7c7b\u522b\u3002b \u6536\u96c6\u73b0\u6709\u7684 58 \u4e2a\u516c\u5171\u6570\u636e\u96c6\u3002c \u5c06\u4e0d\u540c\u6570\u636e\u96c6\u7684\u683c\u5f0f\u8f6c\u6362\u4e3a\u7edf\u4e00\u7684\u4fbf\u4e8e\u6307\u4ee4\u5fae\u8c03\u7684\u683c\u5f0f\u3002d \u6536\u96c6\u7684 MedS-Ins \u7684\u6700\u7ec8\u6570\u636e\u5206\u5e03\u3002\u6851\u57fa\u56fe\u663e\u793a\u4e86\u4e0d\u540c\u6587\u672c\u9886\u57df\uff08\u5de6\u4fa7\uff09\u3001\u4efb\u52a1\u7c7b\u522b\uff08\u4e2d\u95f4\uff09\u548c\u6570\u636e\u6765\u6e90\uff08\u53f3\u4fa7\uff09\u5982\u4f55\u8d21\u732e\u4e8e\u6700\u7ec8\u6570\u636e\u96c6\u3002\u5728\u5e95\u90e8\u5de6\u4fa7\uff0c\u4e24\u4e2a\u997c\u56fe\u5206\u522b\u663e\u793a\u4e86\u6587\u672c\u9886\u57df\u548c\u4efb\u52a1\u7c7b\u522b\u7684\u6570\u636e\u5206\u5e03\u60c5\u51b5\u3002<\/section>\n

\u6587\u672c\u6e90<\/strong><\/p>\n

\u672c\u6587\u63d0\u51fa\u7684\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u96c6\u7531\u4e94\u4e2a\u4e0d\u540c\u6765\u6e90\u7684\u6837\u672c\u7ec4\u6210\uff1a\u8003\u8bd5\u3001\u4e34\u5e8a\u6587\u672c\u3001\u5b66\u672f\u8bba\u6587\u3001\u533b\u5b66\u77e5\u8bc6\u5e93\u548c\u65e5\u5e38\u5bf9\u8bdd\u3002<\/p>\n

\u8003\u8bd5\uff1a<\/strong>\u8be5\u7c7b\u522b\u5305\u542b\u6765\u81ea\u4e0d\u540c\u56fd\u5bb6\u533b\u5b66\u8003\u8bd5\u8bd5\u9898\u7684\u6570\u636e\u3002\u5b83\u6db5\u76d6\u4e86\u4ece\u57fa\u672c\u533b\u5b66\u5e38\u8bc6\u5230\u590d\u6742\u4e34\u5e8a\u624b\u7eed\u5e7f\u6cdb\u7684\u533b\u5b66\u77e5\u8bc6\u3002\u8003\u8bd5\u9898\u76ee\u662f\u4e86\u89e3\u548c\u8bc4\u4f30\u533b\u5b66\u6559\u80b2\u6c34\u5e73\u7684\u91cd\u8981\u624b\u6bb5\uff0c\u7136\u800c\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u8003\u8bd5\u7684\u9ad8\u5ea6\u6807\u51c6\u5316\u5f80\u5f80\u5bfc\u81f4\u5176\u6848\u4f8b\u4e0e\u771f\u5b9e\u4e16\u754c\u7684\u4e34\u5e8a\u4efb\u52a1\u76f8\u6bd4\u8fc7\u4e8e\u7b80\u5316\u3002\u6570\u636e\u96c6\u4e2d 7% \u7684\u6570\u636e\u6765\u81ea\u8003\u8bd5\u3002<\/p>\n

\u4e34\u5e8a\u6587\u672c<\/strong>\uff1a<\/strong>\u8be5\u7c7b\u522b\u6587\u672c\u5728\u5e38\u89c4\u4e34\u5e8a\u5b9e\u8df5\u4e2d\u4ea7\u751f\uff0c\u5305\u62ec\u533b\u9662\u548c\u4e34\u5e8a\u4e2d\u5fc3\u7684\u8bca\u65ad\u3001\u6cbb\u7597\u548c\u9884\u9632\u8fc7\u7a0b\u3002\u8fd9\u7c7b\u6587\u672c\u5305\u62ec\u7535\u5b50\u5065\u5eb7\u8bb0\u5f55 (EHR)\u3001\u653e\u5c04\u62a5\u544a\u3001\u5316\u9a8c\u7ed3\u679c\u3001\u968f\u8bbf\u6307\u5bfc\u548c\u7528\u836f\u5efa\u8bae\u7b49\u3002\u8fd9\u4e9b\u6587\u672c\u662f\u75be\u75c5\u8bca\u65ad\u548c\u60a3\u8005\u7ba1\u7406\u6240\u4e0d\u53ef\u6216\u7f3a\u7684\uff0c\u56e0\u6b64\u51c6\u786e\u7684\u5206\u6790\u548c\u7406\u89e3\u5bf9\u4e8e LLM \u7684\u6709\u6548\u4e34\u5e8a\u5e94\u7528\u81f3\u5173\u91cd\u8981\u3002\u6570\u636e\u96c6\u4e2d 35% \u7684\u6570\u636e\u6765\u81ea\u4e34\u5e8a\u6587\u672c\u3002<\/p>\n

\u5b66\u672f\u8bba\u6587\uff1a<\/strong>\u8be5\u7c7b\u522b\u6570\u636e\u6765\u6e90\u4e8e\u533b\u5b66\u7814\u7a76\u8bba\u6587\uff0c\u6db5\u76d6\u4e86\u533b\u5b66\u7814\u7a76\u9886\u57df\u7684\u6700\u65b0\u53d1\u73b0\u548c\u8fdb\u5c55\u3002\u7531\u4e8e\u5b66\u672f\u8bba\u6587\u4fbf\u4e8e\u83b7\u53d6\u548c\u7ed3\u6784\u5316\u7ec4\u7ec7\uff0c\u4ece\u5b66\u672f\u8bba\u6587\u4e2d\u63d0\u53d6\u6570\u636e\u76f8\u5bf9\u7b80\u5355\u3002\u8fd9\u4e9b\u6570\u636e\u6709\u52a9\u4e8e\u6a21\u578b\u638c\u63e1\u6700\u524d\u6cbf\u7684\u533b\u5b66\u7814\u7a76\u4fe1\u606f\uff0c\u5f15\u5bfc\u6a21\u578b\u66f4\u597d\u5730\u7406\u89e3\u5f53\u4ee3\u533b\u5b66\u7684\u53d1\u5c55\u3002\u6570\u636e\u96c6\u4e2d\u6709 13% \u7684\u6570\u636e\u6765\u81ea\u5b66\u672f\u8bba\u6587\u3002<\/p>\n

\u533b\u5b66\u77e5\u8bc6\u5e93\uff1a<\/strong>\u8be5\u7c7b\u522b\u6570\u636e\u7531\u7ec4\u7ec7\u826f\u597d\u7684\u7efc\u5408\u533b\u5b66\u77e5\u8bc6\u7ec4\u6210\uff0c\u5305\u62ec\u533b\u5b66\u767e\u79d1\u5168\u4e66\u3001\u77e5\u8bc6\u56fe\u8c31\u548c\u533b\u5b66\u672f\u8bed\u8bcd\u6c47\u8868\u3002\u8fd9\u4e9b\u6570\u636e\u6784\u6210\u4e86\u533b\u5b66\u77e5\u8bc6\u5e93\u7684\u6838\u5fc3\uff0c\u4e3a\u533b\u5b66\u6559\u80b2\u548c LLM \u5728\u4e34\u5e8a\u5b9e\u8df5\u4e2d\u7684\u5e94\u7528\u63d0\u4f9b\u4e86\u652f\u6301\u3002\u6570\u636e\u96c6\u4e2d 43% \u7684\u6570\u636e\u6765\u81ea\u533b\u5b66\u77e5\u8bc6\u3002<\/p>\n

\u65e5\u5e38\u5bf9\u8bdd\uff1a<\/strong>\u8be5\u7c7b\u522b\u6570\u636e\u6307\u7684\u662f\u533b\u751f\u4e0e\u60a3\u8005\u4e4b\u95f4\u4ea7\u751f\u7684\u65e5\u5e38\u54a8\u8be2\u3001\u4e3b\u8981\u6765\u6e90\u4e8e\u5728\u7ebf\u5e73\u53f0\u548c\u5176\u4ed6\u4e92\u52a8\u573a\u666f\u3002\u8fd9\u4e9b\u6570\u636e\u53cd\u6620\u4e86\u533b\u52a1\u4eba\u5458\u4e0e\u60a3\u8005\u4e4b\u95f4\u7684\u771f\u5b9e\u4e92\u52a8\u3001\u5728\u4e86\u89e3\u60a3\u8005\u9700\u6c42\u3001\u63d0\u5347\u6574\u4f53\u533b\u7597\u670d\u52a1\u4f53\u9a8c\u65b9\u9762\u53d1\u6325\u7740\u81f3\u5173\u91cd\u8981\u7684\u4f5c\u7528\u3002\u6570\u636e\u96c6\u4e2d\u6709 2% \u7684\u6570\u636e\u6765\u81ea\u65e5\u5e38\u5bf9\u8bdd\u3002<\/p>\n

\u4efb\u52a1\u79cd\u7c7b<\/strong><\/p>\n

\u9664\u4e86\u5bf9\u6587\u672c\u6d89\u53ca\u9886\u57df\u8fdb\u884c\u5206\u7c7b\u5916\uff0c\u7814\u7a76\u56e2\u961f\u5bf9 MedS-Ins \u4e2d\u6837\u672c\u7684\u4efb\u52a1\u7c7b\u522b\u8fdb\u884c\u8fdb\u4e00\u6b65\u7ec6\u5206\uff1a\u786e\u5b9a\u4e86 19 \u4e2a\u4efb\u52a1\u7c7b\u522b\uff0c\u6bcf\u4e2a\u7c7b\u522b\u90fd\u4ee3\u8868\u4e86\u533b\u5b66\u5927\u8bed\u8a00\u6a21\u578b\u5e94\u5177\u5907\u7684\u5173\u952e\u80fd\u529b\u3002\u901a\u8fc7\u6784\u5efa\u8be5\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u96c6\u5e76\u76f8\u5e94\u5730\u5fae\u8c03\u6a21\u578b\uff0c\u4f7f\u5927\u8bed\u8a00\u6a21\u578b\u5177\u5907\u5904\u7406\u533b\u7597\u5e94\u7528\u6240\u9700\u7684\u591a\u79cd\u80fd\u529b\uff0c\u5177\u4f53\u5982\u56fe 2 \u6240\u793a\u3002<\/p>\n

MedS-Ins \u4e2d\u7684 19 \u4e2a\u4efb\u52a1\u7c7b\u522b\u5305\u62ec\u4f46\u4e0d\u9650\u4e8e MedS-Bench \u57fa\u51c6\u4e2d\u7684 11 \u4e2a\u7c7b\u522b\u3002\u989d\u5916\u7684\u4efb\u52a1\u7c7b\u522b\u6db5\u76d6\u4e86\u533b\u5b66\u9886\u57df\u6240\u5fc5\u9700\u7684\u4e00\u7cfb\u5217\u8bed\u8a00\u548c\u5206\u6790\u4efb\u52a1\uff0c\u5305\u62ec\u610f\u56fe\u8bc6\u522b\u3001\u7ffb\u8bd1\u3001\u5355\u8bcd\u5173\u7cfb\u5206\u7c7b\u3001\u6587\u672c\u68c0\u7d22\u3001\u53e5\u5b50\u6210\u5206\u5206\u6790\u3001\u9519\u8bef\u5019\u9009\u8bcd\u751f\u6210\u3001\u5bf9\u8bdd\u548c\u6587\u672c\u8865\u9f50\uff0c\u800c MCQA \u5219\u6269\u5c55\u4e3a\u4e00\u822c\u7684\u95ee\u7b54\u3002\u4efb\u52a1\u7c7b\u522b\u7684\u591a\u6837\u6027\u2014\u2014\u4ece\u666e\u901a\u95ee\u7b54\u548c\u5bf9\u8bdd\u5230\u5404\u79cd\u4e0b\u6e38\u4e34\u5e8a\u4efb\u52a1\uff0c\u4fdd\u8bc1\u4e86\u5bf9\u533b\u7597\u5e94\u7528\u7684\u5168\u9762\u7406\u89e3\u3002<\/p>\n

\u91cf\u5316\u5bf9\u6bd4<\/strong><\/p>\n

\u7814\u7a76\u56e2\u961f\u5e7f\u6cdb\u5730\u6d4b\u8bd5\u4e86\u73b0\u5b58\u516d\u5927\u4e3b\u6d41\u6a21\u578b\uff08MEDITRON, Mistral, InternLM 2, Llama 3, GPT-4 and Claude-3.5\uff09\u5728\u6bcf\u79cd\u4efb\u52a1\u7c7b\u578b\u4e0a\u7684\u8868\u73b0\uff0c\u9996\u5148\u8ba8\u8bba\u5404\u79cd\u73b0\u6709 LLM \u7684\u6027\u80fd\uff0c\u7136\u540e\u4e0e\u63d0\u51fa\u7684\u6700\u7ec8\u6a21\u578b MMedIns-Llama 3 \u8fdb\u884c\u6bd4\u8f83\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6240\u6709\u7ed3\u679c\u90fd\u662f\u4f7f\u7528 3-shot Prompt \u5f97\u51fa\u7684\u3002\u9664\u4e86\u5728 MCQA \u4efb\u52a1\u4e2d\u4f7f\u7528\u4e86 zero-shot Prompt\uff0c\u4ee5\u4fbf\u4e0e\u4e4b\u524d\u7684\u7814\u7a76\u4fdd\u6301\u4e00\u81f4\u3002\u7531\u4e8e GPT-4 \u548c Claude 3.5 \u7b49\u95ed\u6e90\u6a21\u578b\u4f1a\u4ea7\u751f\u8d39\u7528\uff0c\u53d7\u9650\u4e8e\u6210\u672c\uff0c\u5b9e\u9a8c\u4e2d\u4ec5\u5bf9\u6bcf\u4e2a\u4efb\u52a1\u62bd\u6837 50-100 \u4e2a\u6d4b\u8bd5\u7528\u4f8b\uff0c\u5168\u9762\u7684\u6d4b\u8bd5\u91cf\u5316\u7ed3\u679c\u5982\u8868 2-8 \u6240\u793a\u3002<\/p>\n

\u591a\u8bed\u79cd\u591a\u9009\u9898\u95ee\u7b54\uff1a<\/strong>\u8868 2 \u4ee5\u300cAccuracy\u300d\u5c55\u793a\u4e86\u5728\u5e7f\u6cdb\u4f7f\u7528\u7684 MCQA \u57fa\u51c6\u4e0a\u7684\u8bc4\u4f30\u7ed3\u679c\u3002\u5728\u8fd9\u4e9b\u591a\u9009\u9898\u95ee\u7b54\u6570\u636e\u96c6\u4e0a\uff0c\u73b0\u6709\u7684\u5927\u8bed\u8a00\u6a21\u578b\u90fd\u8868\u73b0\u51fa\u4e86\u975e\u5e38\u9ad8\u7684\u51c6\u786e\u7387\uff0c\u4f8b\u5982\uff0c\u5728 MedQA \u4e0a\uff0cGPT-4 \u53ef\u4ee5\u8fbe\u5230 85.8 \u5206\uff0c\u51e0\u4e4e\u53ef\u4ee5\u4e0e\u4eba\u7c7b\u4e13\u5bb6\u76f8\u5ab2\u7f8e\uff0c\u800c Llama 3 \u4e5f\u80fd\u4ee5 60.9 \u5206\u901a\u8fc7\u8003\u8bd5\u3002\u540c\u6837\uff0c\u5728\u82f1\u8bed\u4ee5\u5916\u7684\u8bed\u8a00\u65b9\u9762\uff0cLLM \u5728 MMedBench \u4e0a\u7684\u591a\u9009\u51c6\u786e\u7387\u4e5f\u8868\u73b0\u51fa\u4f18\u5f02\u7684\u6210\u7ee9\u3002<\/p>\n

\u7ed3\u679c\u8868\u660e\uff0c\u7531\u4e8e\u591a\u9009\u9898\u5728\u73b0\u6709\u7814\u7a76\u4e2d\u5df2\u88ab\u5e7f\u6cdb\u8003\u8651\uff0c\u4e0d\u540c\u7684 LLM \u53ef\u80fd\u5df2\u9488\u5bf9\u6b64\u7c7b\u4efb\u52a1\u8fdb\u884c\u4e86\u4e13\u95e8\u4f18\u5316\uff0c\u4ece\u800c\u83b7\u5f97\u4e86\u8f83\u9ad8\u7684\u6027\u80fd\u3002\u56e0\u6b64\uff0c\u6709\u5fc5\u8981\u5efa\u7acb\u4e00\u4e2a\u66f4\u5168\u9762\u7684\u57fa\u51c6\u3001 \u4ee5\u8fdb\u4e00\u6b65\u63a8\u52a8 LLM \u5411\u4e34\u5e8a\u5e94\u7528\u53d1\u5c55\u3002<\/p>\n

\u88682\uff1a\u9009\u62e9\u9898\u4e0a\u7684\u91cf\u5316\u7ed3\u679c\uff0c\u5404\u9879\u6307\u6807\u4ee5\u9009\u62e9\u51c6\u786e\u7387ACC\u8fdb\u884c\u8861\u91cf\u3002<\/p>\n

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

\u6587\u672c\u603b\u7ed3\uff1a<\/strong>\u8868 3 \u4ee5 \u300cBLEU\/ROUGE \u300d\u5206\u6570\u7684\u5f62\u5f0f\u62a5\u544a\u4e86\u4e0d\u540c\u8bed\u8a00\u6a21\u578b\u5728\u6587\u672c\u603b\u7ed3\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\u3002\u6d4b\u8bd5\u8986\u76d6\u4e86\u591a\u79cd\u62a5\u544a\u7c7b\u578b\uff0c\u5305\u62ec X \u5149\u3001CT\u3001MRI\u3001\u8d85\u58f0\u6ce2\u548c\u5176\u4ed6\u533b\u7597\u95ee\u9898\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cGPT-4 \u548c Claude-3.5 \u7b49\u95ed\u6e90\u5927\u8bed\u8a00\u6a21\u578b\u7684\u8868\u73b0\u4f18\u4e8e\u6240\u6709\u5f00\u6e90\u5927\u8bed\u8a00\u6a21\u578b\u3002<\/p>\n

\u5728\u5f00\u6e90\u6a21\u578b\u4e2d\uff0cMistral \u7684\u7ed3\u679c\u6700\u597d\uff0cBLEU\/ROUGE \u5206\u522b\u4e3a 24.48\/24.90\uff0cLlama 3 \u7d27\u968f\u5176\u540e\uff0c\u4e3a 22.20\/23.08\u3002<\/p>\n

\u672c\u6587\u63d0\u51fa\u7684 MMedIns-Llama 3 \u662f\u5728\u7279\u5b9a\u533b\u7597\u6559\u5b66\u6570\u636e\u96c6\uff08MedS-Ins\uff09\u4e0a\u8bad\u7ec3\u51fa\u6765\u7684\uff0c\u5176\u8868\u73b0\u660e\u663e\u4f18\u4e8e\u5176\u4ed6\u6a21\u578b\uff0c\u5305\u62ec\u5148\u8fdb\u7684\u95ed\u6e90\u6a21\u578b GPT-4 \u548c Claude-3.5\uff0c\u5e73\u5747\u5f97\u5206\u8fbe\u5230 46.82\/48.38\u3002<\/p>\n

\u8868 3\uff1a\u6587\u672c\u603b\u7ed3\u4efb\u52a1\u4e0a\u7684\u91cf\u5316\u7ed3\u679c\u3002<\/p>\n

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

\u4fe1\u606f\u62bd\u53d6\uff1a<\/strong>\u8868 4 \u4ee5\u300cAccuracy\u300d\u5c55\u793a\u4e86\u4e0d\u540c\u6a21\u578b\u4fe1\u606f\u63d0\u53d6\u7684\u6027\u80fd\u3002InternLM 2 \u5728\u8fd9\u9879\u4efb\u52a1\u4e2d\u8868\u73b0\u4f18\u5f02\uff0c\u5e73\u5747\u5f97\u5206\u4e3a 81.58\uff0cGPT-4 \u548c Claude-3.5 \u7b49\u95ed\u6e90\u6a21\u578b\u7684\u5e73\u5747\u5f97\u5206\u5206\u522b\u4e3a 77.49 \u5206\u548c 78.86 \u5206\uff0c\u4f18\u4e8e\u6240\u6709\u5176\u4ed6\u5f00\u6e90\u6a21\u578b\u3002<\/p>\n

\u5bf9\u5355\u4e2a\u4efb\u52a1\u7ed3\u679c\u7684\u5206\u6790\u8868\u660e\uff0c\u4e0e\u4e13\u4e1a\u7684\u533b\u7597\u6570\u636e\u76f8\u6bd4\uff0c\u5927\u591a\u6570\u5927\u8bed\u8a00\u6a21\u578b\u5728\u63d0\u53d6\u75c5\u4eba\u57fa\u672c\u4fe1\u606f\u7b49\u4e0d\u592a\u590d\u6742\u7684\u533b\u7597\u4fe1\u606f\u65b9\u9762\u8868\u73b0\u66f4\u597d\u3002\u4f8b\u5982\uff0c\u5728\u4ece PMC \u60a3\u8005\u4e2d\u63d0\u53d6\u57fa\u672c\u4fe1\u606f\u65b9\u9762\uff0c\u5927\u591a\u6570\u5927\u8bed\u8a00\u6a21\u578b\u7684\u5f97\u5206\u90fd\u5728 90 \u5206\u4ee5\u4e0a\uff0c\u5176\u4e2d Claude-3.5 \u7684\u5f97\u5206\u6700\u9ad8\uff0c\u8fbe\u5230 98.02 \u5206\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0cPICO \u4e2d\u4e34\u5e8a\u7ed3\u679c\u63d0\u53d6\u4efb\u52a1\u7684\u8868\u73b0\u76f8\u5bf9\u8f83\u5dee\u3002\u672c\u6587\u63d0\u51fa\u7684\u6a21\u578b MMedIns-Llama 3 \u6574\u4f53\u8868\u73b0\u6700\u4f73\uff0c\u5e73\u5747\u5f97\u5206 83.18\uff0c\u8d85\u8fc7 InternLM 2 \u6a21\u578b 1.6 \u5206\u3002<\/p>\n

\u8868 4\uff1a\u4fe1\u606f\u63d0\u53d6\u4efb\u52a1\u4e0a\u7684\u91cf\u5316\u7ed3\u679c\uff0c\u5404\u9879\u6307\u6807\u4ee5\u51c6\u786e\u5ea6\uff08ACC\uff09\u8fdb\u884c\u8861\u91cf\u3002\u300cExt.\u300d\u8868\u793aExtraction\uff0c\u300cInfo.\u300d\u8868\u793a Information\u3002<\/p>\n

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

\u533b\u5b66\u6982\u5ff5\u89e3\u91ca\uff1a<\/strong>\u8868 5 \u4ee5 \u300cBLEU\/ROUGE \u300d\u5206\u6570\u7684\u5f62\u5f0f\u5c55\u793a\u4e86\u4e0d\u540c\u6a21\u578b\u533b\u5b66\u6982\u5ff5\u89e3\u91ca\u80fd\u529b\uff0cGPT-4\uff0cLlama 3\u548cMistral \u5728\u8fd9\u9879\u4efb\u52a1\u4e2d\u8868\u73b0\u826f\u597d\u3002<\/p>\n

\u76f8\u53cd\uff0cClaude-3.5\u3001InternLM 2 \u548c MEDITRON \u7684\u5f97\u5206\u76f8\u5bf9\u8f83\u4f4e\u3002MEDITRON\u00a0\u7684\u8868\u73b0\u76f8\u5bf9\u8f83\u5dee\u53ef\u80fd\u662f\u7531\u4e8e\u5176\u8bad\u7ec3\u8bed\u6599\u66f4\u4fa7\u91cd\u4e8e\u5b66\u672f\u8bba\u6587\u548c\u6307\u5357\uff0c\u56e0\u6b64\u5728\u5bf9\u4e8e\u533b\u5b66\u6982\u5ff5\u89e3\u91ca\u65b9\u9762\u80fd\u529b\u6709\u6240\u6b20\u7f3a\u3002<\/p>\n

\u6700\u7ec8\u6a21\u578b\u00a0MMedIns-Llama 3\u00a0\u5728\u6240\u6709\u6982\u5ff5\u89e3\u91ca\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u90fd\u660e\u663e\u4f18\u4e8e\u5176\u4ed6\u6a21\u578b\u3002<\/p>\n

\u8868 5\uff1a\u533b\u5b66\u6982\u5ff5\u89e3\u91ca\u4e0a\u7684\u91cf\u5316\u7ed3\u679c\uff0c\u5404\u9879\u6307\u6807\u4ee5 BLEU-1\/ROUGE-1 \u8fdb\u884c\u8861\u91cf\uff1b\u300cExp.\u300d\u8868\u793a Explanation\u3002<\/p>\n

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

\u5f52\u56e0\u5206\u6790\uff08Rationale\uff09\uff1a<\/strong>\u8868 6 \u4ee5 \u300cBLEU\/ROUGE \u300d\u5206\u6570\u7684\u5f62\u5f0f\u8bc4\u4f30\u4e86\u5404\u4e2a\u6a21\u578b\u5728\u5f52\u56e0\u5206\u6790\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\uff0c\u4f7f\u7528 MMedBench \u6570\u636e\u96c6\u5bf9\u516d\u79cd\u8bed\u8a00\u7684\u5404\u79cd\u6a21\u578b\u7684\u63a8\u7406\u80fd\u529b\u8fdb\u884c\u4e86\u6bd4\u8f83\u3002<\/p>\n

\u5728\u6d4b\u8bd5\u7684\u6a21\u578b\u4e2d\uff0c\u95ed\u6e90\u6a21\u578b Claude-3.5 \u8868\u73b0\u51fa\u6700\u5f3a\u7684\u6027\u80fd\uff0c\u5e73\u5747\u5f97\u5206\u4e3a \u00a046.03\/37.65\u3002\u8fd9\u79cd\u4f18\u5f02\u7684\u8868\u73b0\u53ef\u80fd\u662f\u56e0\u4e3a\u8be5\u4efb\u52a1\u4e0e\u751f\u6210\u00a0COT<\/span> \u76f8\u4f3c\uff0c\u800c\u540e\u8005\u5728\u8bb8\u591a\u901a\u7528 LLM \u4e2d\u5747\u5f97\u5230\u4e86\u7279\u522b\u589e\u5f3a\u3002<\/p>\n

\u5728\u5f00\u6e90\u6a21\u578b\u4e2d\uff0cMistral \u548c InternLM 2 \u8868\u73b0\u51fa\u4e86\u76f8\u5f53\u7684\u6027\u80fd\uff0c\u5e73\u5747\u5f97\u5206\u5206\u522b\u4e3a 37.61\/31.55 \u548c 30.03\/26.44\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0cGPT-4 \u88ab\u6392\u9664\u5728\u672c\u6b21\u8bc4\u4f30\u4e4b\u5916\uff0c\u56e0\u4e3a MMedBench \u6570\u636e\u96c6\u7684\u5f52\u56e0\u5206\u6790\u90e8\u5206\u4e3b\u8981\u4f7f\u7528 GPT-4 \u6765\u751f\u6210\u6784\u5efa\uff0c\u8fd9\u53ef\u80fd\u4f1a\u5f15\u5165\u6d4b\u8bd5\u504f\u5dee\uff0c\u4ece\u800c\u5bfc\u81f4\u4e0d\u516c\u5e73\u7684\u6bd4\u8f83\u3002<\/p>\n

\u4e0e\u6982\u5ff5\u89e3\u91ca\u4efb\u52a1\u4e0a\u7684\u8868\u73b0\u4e00\u81f4\uff0c\u6700\u7ec8\u6a21\u578b MMedIns-Llama 3 \u4e5f\u5c55\u73b0\u4e86\u6700\u4f73\u7684\u6574\u4f53\u6027\u80fd\uff0c\u6240\u6709\u8bed\u8a00\u4e0a\u7684\u5e73\u5747\u5f97\u5206\u4e3a 47.17\/34.96\u3002\u8fd9\u79cd\u4f18\u5f02\u7684\u8868\u73b0\u53ef\u80fd\u662f\u56e0\u4e3a\u9009\u7528\u7684\u57fa\u7840\u8bed\u8a00\u6a21\u578b\uff08MMed-Llama 3\uff09\u6700\u521d\u662f\u4e3a\u591a\u8bed\u8a00\u5f00\u53d1\u7684\u3002\u56e0\u6b64\uff0c\u5373\u4f7f\u6307\u4ee4\u8c03\u6574\u6ca1\u6709\u660e\u786e\u9488\u5bf9\u591a\u8bed\u8a00\u6570\u636e\uff0c\u6700\u7ec8\u6a21\u578b\u5728\u591a\u79cd\u8bed\u8a00\u4e2d\u7684\u8868\u73b0\u4ecd\u7136\u4f18\u4e8e\u5176\u4ed6\u6a21\u578b\u3002<\/p>\n

\u8868 6\uff1a\u5f52\u56e0\u5206\u6790\uff08Rationale\uff09\u4e0a\u7684\u91cf\u5316\u7ed3\u679c\uff0c\u5404\u9879\u6307\u6807\u4ee5 BLEU-1\/ROUGE-1 \u8fdb\u884c\u8861\u91cf\u3002\u6b64\u5904\u6ca1\u6709 GPT-4 \u662f\u56e0\u4e3a\u539f\u59cb\u6570\u636e\u57fa\u4e8e GPT-4 \u751f\u6210\u7ed3\u679c\u6784\u9020\uff0c\u5b58\u5728\u516c\u5e73\u6027\u504f\u501a\uff0c\u6545\u672a\u6bd4\u8f83 GPT-4\u3002<\/p>\n

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

\u533b\u5b66\u5b9e\u4f53\u62bd\u53d6\uff08NER\uff09\uff1a<\/strong>\u8868 7 \u4ee5\u300cF1\u300d\u5206\u6570\u7684\u5f62\u5f0f\u6d4b\u8bd5\u4e86\u73b0\u6709\u7684 6 \u4e2a\u6a21\u578b\u5728 NER \u4efb\u52a1\u4e0a\u7684\u8868\u73b0\u3002GPT-4 \u662f\u552f\u4e00\u4e00\u4e2a\u5728\u547d\u540d\u5b9e\u4f53\u8bc6\u522b (NER) \u5404\u9879\u4efb\u52a1\u4e2d\u5747\u8868\u73b0\u4f18\u5f02\u7684\u6a21\u578b\uff0c\u5e73\u5747 F1 \u5206\u6570\u4e3a 44.30 \u3002<\/p>\n

\u5b83\u5728 BC5Chem \u5316\u5b66\u5b9e\u4f53\u8bc6\u522b\u4efb\u52a1\u4e2d\u8868\u73b0\u5c24\u4e3a\u51fa\u8272\uff0c\u5f97\u5206\u4e3a 63.77\u3002InternLM 2 \u5219\u7d27\u968f\u5176\u540e\uff0c\u5e73\u5747 F1 \u5206\u6570\u4e3a 40.81\uff0c\u5728 BC5Chem \u548c BC5Disease \u4efb\u52a1\u4e2d\u5747\u8868\u73b0\u51fa\u8272\u3002Llama 3 \u548c Mistral \u7684\u5e73\u5747 F1 \u5206\u6570\u5219\u5206\u522b\u4e3a 24.70 \u548c 20.10\uff0c\u8868\u73b0\u4e2d\u7b49\u3002MEDITRON \u672a\u9488\u5bf9 NER \u4efb\u52a1\u8fdb\u884c\u4f18\u5316\uff0c\u5728\u6b64\u9886\u57df\u7684\u6548\u679c\u5dee\u5f3a\u4eba\u610f\u3002MMedIns-Llama 3 \u7684\u8868\u73b0\u5219\u660e\u663e\u4f18\u4e8e\u6240\u6709\u5176\u4ed6\u6a21\u578b\uff0c\u5e73\u5747 F1 \u5f97\u5206\u4e3a68.58\u3002<\/p>\n

\u8868 7\uff1aNER \u4efb\u52a1\u4e0a\u7684\u91cf\u5316\u7ed3\u679c\uff0c\u5404\u9879\u6307\u6807\u4ee5F1-score\u8fdb\u884c\u8861\u91cf\uff1b\u300cRec.\u300d\u4ee3\u8868\u300crecognition\u300d<\/p>\n

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

\u8bca\u65ad\u3001\u6cbb\u7597\u63a8\u8350\u3001\u548c\u4e34\u5e8a\u7ed3\u679c\u9884\u6d4b\uff1a<\/strong>\u8868 8 \u4f7f\u7528 DDXPlus \u6570\u636e\u96c6\u4f5c\u4e3a\u8bca\u65ad\u57fa\u51c6\u3001SEER \u6570\u636e\u96c6\u4f5c\u4e3a\u6cbb\u7597\u63a8\u8350\u57fa\u51c6\u548c MIMIC4ED \u6570\u636e\u4f5c\u4e3a\u4e34\u5e8a\u7ed3\u679c\u9884\u6d4b\u4efb\u52a1\u57fa\u51c6\u6765\u8bc4\u4f30\u8bca\u65ad\u3001\u6cbb\u7597\u63a8\u8350\u548c\u4e34\u5e8a\u7ed3\u679c\u9884\u6d4b\u4e09\u5927\u4efb\u52a1\u7684\u6a21\u578b\u8868\u73b0\uff0c\u7ed3\u679c\u4ee5\u51c6\u786e\u5ea6\u6765\u8861\u91cf\uff0c\u5982\u8868 8 \u6240\u793a\u3002<\/p>\n

\u5728\u6b64\uff0c\u4f7f\u7528\u53ef\u4ee5\u4f7f\u7528\u51c6\u786e\u5ea6\u6307\u6807\u6765\u8bc4\u4f30\u751f\u6210\u9884\u6d4b\u662f\u56e0\u4e3a\u8fd9\u4e9b\u6570\u636e\u96c6\u6bcf\u4e00\u4e2a\u90fd\u5c06\u539f\u59cb\u95ee\u9898\u7b80\u5316\u4e3a\u4e00\u4e2a\u95ed\u96c6\u4e0a\u7684\u9009\u62e9\u95ee\u9898\u3002\u5177\u4f53\u800c\u8a00\uff0cDDXPlus \u4f7f\u7528\u9884\u5b9a\u4e49\u7684\u75be\u75c5\u5217\u8868\uff0c\u6a21\u578b\u5fc5\u987b\u6839\u636e\u63d0\u4f9b\u7684\u60a3\u8005\u80cc\u666f\u4ece\u4e2d\u9009\u62e9\u4e00\u4e2a\u75be\u75c5\u3002\u5728 SEER \u4e2d\uff0c\u6cbb\u7597\u5efa\u8bae\u5219\u88ab\u5206\u4e3a\u4e86\u516b\u4e2a\u9ad8\u7ea7\u7c7b\u522b\uff0c\u800c\u5728 MIMIC4ED \u4e2d\uff0c\u6700\u7ec8\u7684\u4e34\u5e8a\u7ed3\u679c\u51b3\u7b56\u662f\u59cb\u7ec8\u662f\u4e8c\u503c\u7684\uff08True or False\uff09\u3002<\/p>\n

\u603b\u4f53\u800c\u8a00\uff0c\u5f00\u6e90 LLM \u5728\u8fd9\u4e9b\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\u4e0d\u5982\u95ed\u6e90 LLM\uff0c\u5728\u67d0\u4e9b\u60c5\u51b5\u4e0b\uff0c\u5b83\u4eec\u65e0\u6cd5\u63d0\u4f9b\u6709\u610f\u4e49\u7684\u9884\u6d4b\u3002\u4f8b\u5982\uff0cLlama 3 \u5728\u9884\u6d4b Critical Triage \u65b9\u9762\u8868\u73b0\u4e0d\u4f73\u3002\u5bf9\u4e8e DDXPlus \u8bca\u65ad\u4efb\u52a1\u800c\u8a00\uff0cInternLM 2 \u548c Llama 3 \u7684\u8868\u73b0\u7565\u597d\u4e00\u4e9b\uff0c\u51c6\u786e\u5ea6\u4e3a 32\u3002\u7136\u800c\uff0cGPT-4 \u548c Claude-3.5 \u7b49\u95ed\u6e90\u6a21\u578b\u8868\u73b0\u51fa\u660e\u663e\u66f4\u597d\u7684\u6027\u80fd\u3002\u4f8b\u5982\uff0cClaude-3.5\u5728SEER\u4e0a\u51c6\u786e\u5ea6\u53ef\u4ee5\u8fbe\u5230\u4e3a90\uff0c\u800cGPT-4\u5219\u5728 DDXPlus \u7684\u8bca\u65ad\u65b9\u9762\u7684\u51c6\u786e\u5ea6\u4e0a\u66f4\u9ad8\uff0c\u5f97\u5206\u4e3a 52\uff0c\u7a81\u663e\u51fa\u4e86\u5f00\u6e90\u548c\u95ed\u6e90 LLM \u4e4b\u95f4\u7684\u5de8\u5927\u5dee\u8ddd\u3002<\/p>\n

\u5c3d\u7ba1\u53d6\u5f97\u4e86\u8fd9\u4e9b\u6210\u679c\uff0c\u4f46\u8fd9\u4e9b\u5206\u6570\u4ecd\u7136\u4e0d\u8db3\u4ee5\u53ef\u9760\u5730\u7528\u4e8e\u4e34\u5e8a\u3002\u76f8\u6bd4\u4e4b\u4e0b\uff0c MMedIns-Llama 3 \u5728\u4e34\u5e8a\u51b3\u7b56\u652f\u6301\u4efb\u52a1\u4e2d\u5219\u8868\u73b0\u51fa\u4e86\u66f4\u52a0\u5353\u8d8a\u7684\u51c6\u786e\u6027\uff0c\u4f8b\u5982 SEER \u4e0a\u4e3a 98\uff0cDDXPlus \u4e0a\u4e3a 95\uff0c\u4e34\u5e8a\u7ed3\u679c\u9884\u6d4b\u4efb\u52a1\u4e0a\u5e73\u5747\u51c6\u786e\u5ea6\u4e3a86.67\uff08Hospitalization, 72h ED Revisit, and Critical Triage \u5f97\u5206\u7684\u5e73\u5747\u503c\uff09\u3002<\/p>\n

\u6587\u672c\u5206\u7c7b\uff1a<\/strong>\u8868 8 \u8fd8\u5c55\u793a\u4e86\u5bf9\u00a0HoC<\/span> \u591a\u6807\u7b7e\u5206\u7c7b\u4efb\u52a1\u7684\u8bc4\u4f30\uff0c\u5e76\u62a5\u544a\u4e86 Macro-precision\u3001Macro-recall \u548c Macro-F1 Scores\u3002\u5bf9\u4e8e\u8fd9\u7c7b\u4efb\u52a1\uff0c\u6240\u6709\u5019\u9009\u6807\u7b7e\u90fd\u4ee5\u5217\u8868\u7684\u5f62\u5f0f\u8f93\u5165\u5230\u8bed\u8a00\u6a21\u578b\u4e2d\uff0c\u5e76\u8981\u6c42\u6a21\u578b\u9009\u62e9\u5176\u5bf9\u5e94\u7684\u7b54\u6848\uff0c\u5e76\u5141\u8bb8\u8fdb\u884c\u591a\u9879\u9009\u62e9\u3002\u7136\u540e\u6839\u636e\u6a21\u578b\u6700\u7ec8\u7684\u9009\u62e9\u8f93\u51fa\u8ba1\u7b97\u51c6\u786e\u5ea6\u6307\u6807\u3002<\/p>\n

GPT-4 \u548c Claude-3.5 \u5728\u6b64\u4efb\u52a1\u4e0a\u8868\u73b0\u826f\u597d\uff0cGPT-4 \u7684 Macro-F1 \u5206\u6570\u4e3a 60.38\uff0cClaude-3.5 \u5219\u66f4\u4e3a\u4f18\u5f02\uff0c\u53d6\u5f97\u4e8663.32\u3002\u8fd9\u4e24\u4e2a\u6a21\u578b\u90fd\u8868\u73b0\u51fa\u5f88\u5f3a\u7684\u53ec\u56de\u80fd\u529b\uff0c\u5c24\u5176\u662f Claude-3.5\uff0c\u5176 Macro-Recall \u4e3a 80.96\u3002Mistral \u8868\u73b0\u4e2d\u7b49\uff0cMacro-F1 \u5206\u6570\u4e3a 40.8\uff0c\u5728\u7cbe\u5ea6\u548c\u53ec\u56de\u7387\u4e4b\u95f4\u4fdd\u6301\u5e73\u8861\u3002<\/p>\n

\u76f8\u6bd4\u4e4b\u4e0b\uff0cLlama 3 \u548c InternLM 2 \u7684\u6574\u4f53\u8868\u73b0\u8f83\u5dee\uff0cMacro-F1 \u5f97\u5206\u5206\u522b\u4e3a36.18 \u548c 32.72\u3002\u8fd9\u4e9b\u6a21\u578b\uff08\u5c24\u5176\u662f InternLM 2\uff09\u8868\u73b0\u51fa\u8f83\u9ad8\u7684\u53ec\u56de\u7387\uff0c\u4f46\u51c6\u786e\u7387\u5374\u5f88\u5dee\uff0c\u5bfc\u81f4 Macro-F1 \u5f97\u5206\u8f83\u4f4e\u3002<\/p>\n

MEDITRON \u5728\u6b64\u4efb\u52a1\u4e2d\u6392\u540d\u6700\u4f4e\uff0cMacro-F1 \u5f97\u5206\u4e3a 26.21\u3002MMedIns-Llama 3 \u660e\u663e\u4f18\u4e8e\u6240\u6709\u5176\u4ed6\u6a21\u578b\uff0c\u5728\u6240\u6709\u6307\u6807\u4e2d\u5747\u83b7\u5f97\u6700\u9ad8\u5206\uff0cMacro-precision \u4e3a 91.29\uff0cMacro-recall \u4e3a 85.57\uff0cMacro-F1 \u5f97\u5206\u4e3a 87.37\u3002\u8fd9\u4e9b\u7ed3\u679c\u51f8\u663e\u4e86 MMedIns-Llama 3 \u51c6\u786e\u5206\u7c7b\u6587\u672c\u7684\u80fd\u529b\uff0c\u4f7f\u5176\u6210\u4e3a\u5904\u7406\u8fd9\u7c7b\u590d\u6742\u4efb\u52a1\u6700\u6709\u6548\u7684\u6a21\u578b\u3002<\/p>\n

\u8868 8\uff1a\u6cbb\u7597\u89c4\u5212\uff08SEER\uff09\u3001\u8bca\u65ad\uff08DDXPlus\uff09\u3001\u4e34\u5e8a\u7ed3\u679c\u9884\u6d4b\uff08MIMIC4ED\uff09\u548c\u6587\u672c\u5206\u7c7b(HoC Classification)\u56db\u7c7b\u4efb\u52a1\u4e0a\u7684\u7ed3\u679c\u3002\u524d 3 \u9879\u4efb\u52a1\u7684\u7ed3\u679c\u4ee5\u51c6\u786e\u7387\uff08Accuracy\uff09\u4e3a\u4f9d\u636e\uff0c\u6587\u672c\u5206\u7c7b\u7ed3\u679c\u4ee5\u7cbe\u786e\u5ea6\uff08Precision\uff09\u3001\u53ec\u56de\u7387\uff08Recall\uff09\u548c F1 \u5206\u6570\u4e3a\u4f9d\u636e\u3002<\/p>\n

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

\u4e8b\u5b9e\u7ea0\u6b63\uff1a<\/strong>\u8868 9 \u5c55\u793a\u4e86\u5728\u4e8b\u5b9e\u9a8c\u8bc1\u4efb\u52a1\u4e0a\u6a21\u578b\u8bc4\u4f30\u7ed3\u679c\u3002\u5bf9\u4e8e PubMedQA \u7b54\u6848\u9a8c\u8bc1\u548c HealthFact \u9a8c\u8bc1\uff0cLLM \u9700\u8981\u4ece\u63d0\u4f9b\u7684\u5019\u9009\u5217\u8868\u4e2d\u9009\u62e9\u4e00\u4e2a\u7b54\u6848\uff0c\u56e0\u6b64\u4ee5\u51c6\u786e\u5ea6\u4f5c\u4e3a\u8bc4\u4f30\u6307\u6807\u3002<\/p>\n

\u76f8\u5bf9\u7684\uff0c\u7531\u4e8e EBMS \u7406\u7531\u9a8c\u8bc1\uff0c\u4efb\u52a1\u6d89\u53ca\u751f\u6210\u81ea\u7531\u683c\u5f0f\u7684\u6587\u672c\uff0c\u4f7f\u7528 BLEU-1 \u548c ROUGE-1 \u5206\u6570\u6765\u8bc4\u4f30\u6027\u80fd\u3002InternLM 2 \u5728 PubMedQA \u7b54\u6848\u9a8c\u8bc1\u548c HealthFact \u9a8c\u8bc1\u4e2d\u83b7\u5f97\u4e86\u6700\u9ad8\u7684\u51c6\u786e\u5ea6\uff0c\u5f97\u5206\u5206\u522b\u4e3a 98 \u548c 92\u3002<\/p>\n

\u5728 EBMS \u57fa\u51c6\u6d4b\u8bd5\u4e2d\uff0cGPT-4 \u8868\u73b0\u51fa\u6700\u5f3a\u7684\u6027\u80fd\uff0cBLEU-1\/ROUGE-1 \u5f97\u5206\u5206\u522b\u4e3a16.36\/16.33\u3002Claude-3.5 \u7d27\u968f\u5176\u540e\uff0c\u5f97\u5206\u4e3a14.22\/15.82\uff0c\u4f46\u5b83\u5728PubMedQA\u7b54\u6848\u9a8c\u8bc1\u4e2d\u8868\u73b0\u4e0d\u4f73\u3002<\/p>\n

Llama 3 \u5728 PubMedQA \u548c HealthFact Verification \u4e0a\u7684\u51c6\u786e\u7387\u5206\u522b\u4e3a 94 \u548c 64\uff0cBLEU-1\/ROUGE-1 \u5f97\u5206\u4e3a12.96\/14.37\u3002MMedIns-Llama 3 \u7ee7\u7eed\u8d85\u8d8a\u73b0\u6709\u6a21\u578b\uff0c\u4e0e InternLM 2 \u4e00\u6837\u5728PubMedQA\u7b54\u6848\u9a8c\u8bc1\u4efb\u52a1\u4e0a\u53d6\u5f97\u4e86\u6700\u9ad8\u7684\u51c6\u786e\u5ea6\u5f97\u5206\uff0c\u800c\u5728 EMBS \u4e2d\uff0cMMedIns-Llama 3 \u5728 BLEU-1\u548c ROUGE-1 \u4e2d\u4ee5 11.99\/12.90 \u7684\u6210\u7ee9\u7565\u5fae\u843d\u540e\u4e8e GPT-4\u3002<\/p>\n

\u533b\u5b66\u6587\u672c\u8574\u542b\uff08NLI\uff09\uff1a<\/strong>\u8868 9 \u8fd8\u5c55\u793a\u4e86\u4ee5 MedNLI \u4e3a\u4e3b\u7684\uff0c\u5728\u533b\u5b66\u6587\u672c\u8574\u542b(NLI)\u4e0a\u7684\u8bc4\u4f30\u7ed3\u679c\u3002\u6d4b\u8bd5\u65b9\u5f0f\u6709\u4e24\u79cd\uff0c\u4e00\u79cd\u662f\u5224\u522b\u4efb\u52a1\uff08\u4ece\u5019\u9009\u5217\u8868\u4e2d\u9009\u62e9\u6b63\u786e\u7b54\u6848\uff09\uff0c\u4ee5\u51c6\u786e\u5ea6\u8861\u91cf\uff0c\u53e6\u4e00\u79cd\u662f\u751f\u6210\u4efb\u52a1\uff08\u751f\u6210\u81ea\u7531\u683c\u5f0f\u6587\u672c\u7b54\u6848\uff09\uff0c\u4ee5BLEU\/ROUGE \u6307\u6807\u6765\u8861\u91cf\u3002<\/p>\n

InternLM 2 \u5728\u5f00\u6e90 LLM \u4e2d\u5f97\u5206\u6700\u9ad8\uff0c\u5f97\u5206\u4e3a 84\u3002\u5bf9\u4e8e\u95ed\u6e90 LLM\uff0cGPT-4 \u548c Claude-3.5 \u90fd\u663e\u793a\u51fa\u76f8\u5bf9\u8f83\u9ad8\u7684\u5206\u6570\uff0c\u51c6\u786e\u5ea6\u5206\u522b\u4e3a 82 \u548c 86\u3002\u5728\u751f\u6210\u4efb\u52a1\u4e2d\uff0cLlama 3 \u4e0e\u771f\u5b9e\u503c\u7684\u4e00\u81f4\u6027\u6700\u9ad8\uff0cBLEU \u548c ROUGE \u5f97\u5206\u4e3a 33.69\/27.28\u3002Mistral \u548c Llama 3 \u5219\u8868\u73b0\u51fa\u4e2d\u7b49\u6c34\u5e73\u3002GPT-4 \u7d27\u968f\u5176\u540e\uff0c\u5f97\u5206\u4e3a 27.09\/23.71\uff0c\u800c Claude-3.5 \u5728\u751f\u6210\u4efb\u52a1\u4e2d\u8868\u73b0\u5e76\u4e0d\u7406\u60f3\u3002<\/p>\n

MMedIns-Llama 3 \u5728\u5224\u522b\u4efb\u52a1\u4e2d\u51c6\u786e\u7387\u6700\u9ad8\uff0c\u5f97\u5206\u4e3a 84\uff0c\u4f46\u7565\u843d\u540e\u4e8e Claude-3.5\u3002MMedIns-Llama 3 \u5728\u751f\u6210\u4efb\u52a1\u4e2d\u4e5f\u8868\u73b0\u51fa\u8272\uff0cBLEU\/ROUGE \u00a0\u5f97\u5206\u4e3a 37.56\/32.17\uff0c\u660e\u663e\u4f18\u4e8e\u5176\u4ed6\u6a21\u578b\u3002<\/p>\n

\u8868 9\uff1a\u4e8b\u5b9e\u9a8c\u8bc1\u548c\u6587\u672c\u8574\u542b\u4e24\u7c7b\u4efb\u52a1\u4e0a\u7684\u91cf\u5316\u7ed3\u679c\uff0c\u7ed3\u679c\u4ee5\u51c6\u786e\u5ea6\uff08ACC\uff09\u548cBLEU\/ROUGE\u6765\u8861\u91cf\uff1b\u8868\u4e2d\u300cVer.\u300d\u662f\u300cverification\u300d\u7684\u7b80\u5199\u3002<\/p>\n

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

\u603b\u7684\u800c\u8a00\uff0c\u7814\u7a76\u56e2\u961f\u5728\u5404\u79cd\u4efb\u52a1\u7ef4\u5ea6\u4e0a\uff0c\u8bc4\u6d4b\u4e86\u516d\u5927\u4e3b\u6d41\u6a21\u578b\uff0c\u7814\u7a76\u7ed3\u679c\u8868\u660e\u76ee\u524d\u7684\u4e3b\u6d41 LLMs \u5904\u7406\u4e34\u5e8a\u4efb\u52a1\u65f6\u4ecd\u65e7\u76f8\u5f53\u8106\u5f31\uff0c\u4f1a\u5728\u591a\u6837\u5316\u7684\u590d\u6742\u4e34\u5e8a\u573a\u666f\u4e0b\u4ea7\u751f\u4e25\u91cd\u7684\u6027\u80fd\u4e0d\u8db3\u3002<\/p>\n

\u540c\u65f6\uff0c\u5b9e\u9a8c\u7ed3\u679c\u4e5f\u5c55\u793a\u51fa\uff0c\u901a\u8fc7\u5728\u6307\u4ee4\u6570\u636e\u96c6\u4e2d\u52a0\u5165\u66f4\u591a\u4e34\u5e8a\u4efb\u52a1\u6587\u672c\uff0c\u5f3a\u5316 \u00a0LLM \u4e0e\u4e34\u5e8a\u5b9e\u9645\u9002\u7528\u7684\u5339\u914d\u7a0b\u5ea6\uff0c\u53ef\u4ee5\u5927\u5927\u7684\u52a0\u5f3a LLM \u6027\u80fd\u3002<\/p>\n

\u6570\u636e\u6536\u96c6\u65b9\u6cd5\u4e0e\u8bad\u7ec3\u6d41\u7a0b<\/strong><\/p>\n

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

\u56fe 3\uff1a\u6570\u636e\u6784\u5efa\u6d41\u7a0b\u3002a \u901a\u8fc7\u8fc7\u6ee4\u81ea\u7136\u6307\u4ee4\u548c\u63d0\u793a\u7ec4\u7ec7\u826f\u597d\u7684 BioNLP \u6570\u636e\u96c6\u6765\u6536\u96c6\u6570\u636e\u3002b \u5229\u7528\u6536\u96c6\u5230\u7684 MedS-Ins \u5bf9\u6a21\u578b\u8fdb\u884c\u8bad\u7ec3\u548c\u8bc4\u4f30\u7684\u6d41\u7a0b\u3002\u5229\u7528\u6307\u4ee4\u8c03\u6574\u8bad\u7ec3\u65b9\u6cd5\u7ed3\u5408\u4e0d\u540c\u7684\u6570\u636e\u96c6\uff0c\u5e76\u5728\u591a\u4e2a\u57fa\u51c6\u4e0a\u5bf9\u6700\u7ec8\u6a21\u578b\u8fdb\u884c\u7efc\u5408\u8bc4\u4f30\u3002<\/section>\n

\u672c\u8282\u5c06\u8be6\u7ec6\u4ecb\u7ecd\u8bad\u7ec3\u8fc7\u7a0b\uff0c\u5982\u56fe 3b \u6240\u793a\u3002\u5177\u4f53\u65b9\u6cd5\u4e0e\u4e4b\u524d\u7684\u5de5\u4f5c MMedLM \u548c PMC-LLaMA \u76f8\u540c\uff0c\u5747\u901a\u8fc7\u5728\u533b\u5b66\u76f8\u5173\u7684\u8bed\u6599\u5e93\u4e0a\u8fdb\u884c\u8fdb\u4e00\u6b65\u7684\u81ea\u56de\u5f52\u8bad\u7ec3\u53ef\u4ee5\u4e3a\u6a21\u578b\u6ce8\u5165\u76f8\u5e94\u7684\u533b\u5b66\u77e5\u8bc6\uff0c\u4ece\u800c\u4f7f\u5b83\u4eec\u5728\u4e0d\u540c\u7684\u4e0b\u6e38\u4efb\u52a1\u4e2d\u8868\u73b0\u5f97\u66f4\u597d\u3002<\/p>\n

\u5177\u4f53\u800c\u8a00\uff0c\u7814\u7a76\u56e2\u961f\u4ece\u591a\u8bed\u8a00 LLM \u57fa\u5ea7\u6a21\u578b\uff08MMed-Llama 3\uff09\u5f00\u59cb\uff0c\u5229\u7528\u6765\u81ea MedS-Ins \u7684\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u5bf9\u5176\u8fdb\u884c\u8fdb\u4e00\u6b65\u8bad\u7ec3\u3002<\/p>\n

\u6307\u4ee4\u5fae\u8c03\u7684\u6570\u636e\u4e3b\u8981\u6d89\u53ca\u4e24\u4e2a\u65b9\u9762\uff1a<\/p>\n

\u533b\u5b66\u8fc7\u6ee4\u540e\u7684\u81ea\u7136\u6307\u4ee4\u6570\u636e\uff1a<\/strong>\u9996\u5148\u4ece\u81ea\u7136\u9886\u57df\u6700\u5927\u89c4\u6a21\u7684\u6307\u4ee4\u6570\u636e\u96c6Super-NaturalInstructions\u4e2d\u7b5b\u9009\u51fa\u4e0e\u533b\u7597\u76f8\u5173\u7684\u4efb\u52a1\u3002\u7531\u4e8eSuper-NaturalInstructions\u66f4\u4fa7\u91cd\u4e8e\u901a\u7528\u9886\u57df\u7684\u4e0d\u540c\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4efb\u52a1\uff0c\u56e0\u6b64\u5bf9\u533b\u7597\u9886\u57df\u7684\u5206\u7c7b\u7c92\u5ea6\u76f8\u5bf9\u8f83\u7c97\u3002<\/p>\n

\u9996\u5148\u63d0\u53d6\u4e86 \u300c\u533b\u7597\u4fdd\u5065 \u300d\u548c \u300c\u533b\u5b66 \u300d\u7c7b\u522b\u4e2d\u7684\u6240\u6709\u6307\u4ee4\uff0c\u7136\u540e\u5728\u4efb\u52a1\u7c7b\u522b\u4e0d\u53d8\u7684\u60c5\u51b5\u4e0b\uff0c\u624b\u52a8\u4e3a\u5b83\u4eec\u6dfb\u52a0\u4e86\u66f4\u8be6\u7ec6\u7684\u9886\u57df\u6807\u7b7e\u3002\u6b64\u5916\uff0c\u8bb8\u591a\u901a\u7528\u9886\u57df\u7684\u6709\u7ec4\u7ec7\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u96c6\u4e5f\u6db5\u76d6\u4e86\u4e00\u4e9b\u533b\u7597\u76f8\u5173\u6570\u636e\uff0c\u4f8b\u5982LIMA\u548cShareGPT\u3002<\/p>\n

\u4e3a\u4e86\u8fc7\u6ee4\u51fa\u8fd9\u4e9b\u6570\u636e\u4e2d\u7684\u533b\u7597\u90e8\u5206\uff0c\u7814\u7a76\u56e2\u961f\u4f7f\u7528\u4e86InsTag\u5bf9\u6bcf\u6761\u6307\u4ee4\u7684\u9886\u57df\u8fdb\u884c\u7c97\u7c92\u5ea6\u5206\u7c7b\u3002\u5177\u4f53\u6765\u8bf4\uff0cInsTag\u662f\u4e00\u79cd LLM\uff0c\u4e13\u95e8\u7528\u4e8e\u6807\u8bb0\u4e0d\u540c\u7684\u6307\u4ee4\u6837\u672c\u3002\u7ed9\u5b9a\u4e00\u4e2a\u6307\u4ee4\u67e5\u8be2\uff0c\u5b83\u5c06\u5206\u6790\u8be5\u6307\u4ee4\u5c5e\u4e8e\u54ea\u4e2a\u9886\u57df\u548c\u4efb\u52a1\uff0c\u5728\u6b64\u57fa\u7840\u4e0a\u7b5b\u9009\u51fa\u6807\u8bb0\u4e3a\u533b\u7597\u4fdd\u5065\u3001\u533b\u7597\u6216\u751f\u7269\u533b\u5b66\u7684\u6837\u672c\u3002<\/p>\n

\u6700\u540e\uff0c\u901a\u8fc7\u8fc7\u6ee4\u4e00\u822c\u9886\u57df\u7684\u6307\u4ee4\u6570\u636e\u96c6\uff0c\u6536\u96c6\u5230 37 \u4e2a\u4efb\u52a1\uff0c\u5171\u8ba1 75373 \u4e2a\u6837\u672c\u3002<\/p>\n

\u63d0\u793a\u6784\u9020\u73b0\u6709\u7684 BioNLP \u6570\u636e\u96c6\uff1a<\/strong>\u5728\u73b0\u6709\u6570\u636e\u96c6\u4e2d\u6709\u8bb8\u591a\u5173\u4e8e\u4e34\u5e8a\u573a\u666f\u4e2d\u6587\u672c\u5206\u6790\u7684\u4f18\u79c0\u6570\u636e\u96c6\u3002\u7136\u800c\uff0c\u7531\u4e8e\u5927\u591a\u6570\u6570\u636e\u96c6\u7684\u6536\u96c6\u76ee\u7684\u4e0d\u540c\uff0c\u5b83\u4eec\u4e0d\u80fd\u76f4\u63a5\u7528\u4e8e\u8bad\u7ec3\u5927\u578b\u8bed\u8a00\u6a21\u578b\u3002\u7136\u800c\uff0c\u53ef\u4ee5\u901a\u8fc7\u5c06\u8fd9\u4e9b\u73b0\u6709\u7684\u533b\u5b66 NLP \u4efb\u52a1\u8f6c\u6362\u6210\u53ef\u7528\u4e8e\u8bad\u7ec3\u751f\u6210\u6a21\u578b\u7684\u683c\u5f0f\uff0c\u4ece\u800c\u5c06\u5b83\u4eec\u52a0\u5165\u5230\u6307\u4ee4\u8c03\u6574\u4e2d\u3002<\/p>\n

\u5177\u4f53\u6765\u8bf4\uff0c\u7814\u7a76\u56e2\u961f\u4ee5 MIMIC-IV-Note \u4e3a\u4f8b\u3002MIMIC-IV-Note \u63d0\u4f9b\u4e86\u9ad8\u8d28\u91cf\u7684\u7ed3\u6784\u5316\u62a5\u544a\uff0c\u5176\u4e2d\u65e2\u6709\u53d1\u73b0\u4e5f\u6709\u4e5f\u6709\u7ed3\u8bba\uff0c\u53d1\u73b0\u5230\u7ed3\u8bba\u7684\u751f\u6210\u88ab\u89c6\u4e3a\u7ecf\u5178\u7684\u4e34\u5e8a\u6587\u672c\u603b\u7ed3\u4efb\u52a1\u3002\u9996\u5148\u624b\u52a8\u7f16\u5199\u63d0\u793a\u6765\u5b9a\u4e49\u4efb\u52a1\uff0c\u4f8b\u5982\uff1a\u300c\u9274\u4e8e\u8d85\u58f0\u6210\u50cf\u8bca\u65ad\u7684\u8be6\u7ec6\u7ed3\u679c\uff0c\u7528\u51e0\u4e2a\u8bcd\u6982\u62ec\u53d1\u73b0\u3002\u300d \u8003\u8651\u5230\u6307\u4ee4\u8c03\u6574\u7684\u591a\u6837\u6027\u9700\u6c42\uff0c\u7814\u7a76\u56e2\u961f\u8981\u6c425\u4e2a\u4eba\u72ec\u7acb\u5730\u75283\u79cd\u4e0d\u540c\u7684\u63d0\u793a\u63cf\u8ff0\u67d0\u9879\u4efb\u52a1\u3002<\/p>\n

\u8fd9\u6837\uff0c\u6bcf\u4e2a\u4efb\u52a1\u5c31\u6709\u4e86 15 \u6761\u81ea\u7531\u6587\u672c\u63d0\u793a\uff0c\u4fdd\u8bc1\u4e86\u8bed\u4e49\u76f8\u4f3c\uff0c\u4f46\u63aa\u8f9e\u548c\u683c\u5f0f\u5c3d\u53ef\u80fd\u591a\u6837\u5316\u3002\u7136\u540e\uff0c\u53d7 Self-Instruct \u7684\u542f\u53d1\uff0c\u5c06\u8fd9\u4e9b\u4eba\u5de5\u7f16\u5199\u7684\u6307\u4ee4\u4f5c\u4e3a\u79cd\u5b50\u6307\u4ee4\uff0c\u5e76\u8981\u6c42 GPT-4 \u6839\u636e\u5176\u8fdb\u884c\u6539\u5199\uff0c\u4ece\u800c\u83b7\u5f97\u66f4\u591a\u7684\u591a\u6837\u5316\u6307\u4ee4\u3002<\/p>\n

\u901a\u8fc7\u4e0a\u8ff0\u8fc7\u7a0b\uff0c\u5c06\u989d\u5916\u7684 85 \u4e2a\u4efb\u52a1\u63d0\u793a\u4e3a\u7edf\u4e00\u7684\u81ea\u7531\u95ee\u7b54\u683c\u5f0f\uff0c\u518d\u7ed3\u5408\u8fc7\u6ee4\u540e\u7684\u6570\u636e\uff0c\u5f97\u5230\u4e86\u603b\u8ba1 1350 \u4e07\u4e2a\u9ad8\u8d28\u91cf\u6837\u672c\uff0c\u6db5\u76d6 122 \u4e2a\u4efb\u52a1\uff0c\u79f0\u4e3a MedS-Ins\uff0c\u5e76\u901a\u8fc7\u6307\u4ee4\u5fae\u8c03\uff0c\u8bad\u7ec3\u4e86\u4e00\u4e2a\u5168\u65b0\u7684 8B \u5c3a\u5bf8\u7684\u533b\u5b66 LLM\uff0c\u7ed3\u679c\u8868\u660e\u8be5\u65b9\u6cd5\u663e\u8457\u63d0\u9ad8\u4e86\u4e34\u5e8a\u4efb\u52a1\u7684\u6027\u80fd\u3002<\/p>\n

\u5728\u6307\u4ee4\u5fae\u8c03\u4e2d\uff0c\u7814\u7a76\u56e2\u961f\u91cd\u70b9\u8003\u8651\u4e86\u4e24\u79cd\u6307\u4ee4\u5f62\u5f0f\uff1a<\/p>\n

\u96f6\u6837\u672c\u63d0\u793a\uff1a<\/strong>\u5728\u8fd9\u91cc\uff0c\u4efb\u52a1\u7684\u6307\u4ee4\u5305\u542b\u4e00\u4e9b\u8bed\u4e49\u4efb\u52a1\u63cf\u8ff0\u4f5c\u4e3a\u63d0\u793a\uff0c\u56e0\u6b64\u8981\u6c42\u6a21\u578b\u6839\u636e\u5176\u5185\u90e8\u6a21\u578b\u77e5\u8bc6\u76f4\u63a5\u56de\u7b54\u95ee\u9898\u3002\u5728\u6536\u96c6\u5230\u7684 MedS-Ins \u4e2d\uff0c\u6bcf\u4e2a\u4efb\u52a1\u7684 \u300c\u5b9a\u4e49 \u300d\u5185\u5bb9\u90fd\u53ef\u4ee5\u81ea\u7136\u5730\u7528\u4f5c\u96f6\u70b9\u6307\u4ee4\u8f93\u5165\u3002\u7531\u4e8e\u6db5\u76d6\u4e86\u5404\u79cd\u4e0d\u540c\u7684\u533b\u7597\u4efb\u52a1\u5b9a\u4e49\uff0c\u8be5\u6a21\u578b\u6709\u671b\u5b66\u4e60\u5bf9\u5404\u79cd\u4efb\u52a1\u63cf\u8ff0\u7684\u8bed\u4e49\u7406\u89e3\u3002<\/p>\n

\u5c11\u6837\u672c\u63d0\u793a\uff1a<\/strong>\u5728\u8fd9\u91cc\uff0c\u6307\u4ee4\u5305\u542b\u4e86\u5c11\u91cf\u7684\u793a\u4f8b\uff0c\u8fd9\u4e9b\u793a\u4f8b\u5141\u8bb8\u6a21\u578b\u4ece\u4e0a\u4e0b\u6587\u4e2d\u5b66\u4e60\u5230\u4efb\u52a1\u7684\u5927\u81f4\u9700\u6c42\u3002\u53ea\u9700\u4ece\u540c\u4e00\u4efb\u52a1\u7684\u8bad\u7ec3\u96c6\u4e2d\u968f\u673a\u62bd\u53d6\u5176\u4ed6\u6848\u4f8b\uff0c\u5e76\u4f7f\u7528\u4ee5\u4e0b\u7b80\u5355\u6a21\u677f\u5bf9\u5176\u8fdb\u884c\u7ec4\u7ec7\uff0c\u5373\u53ef\u83b7\u5f97\u6b64\u7c7b\u6307\u4ee4\uff1a<\/p>\n

\n
\n
Case1: Input: {CASE1_INPUT}, \u00a0Output: {CASE1_OUTPUT} … CaseN: Input: {CASEN_INPUT}, \u00a0Output: {CASEN_OUTPUT} {INSTRUCTION} Please learn from the few-shot \u00a0cases to see what content you have to output. Input: {INPUT}<\/section>\n<\/section>\n<\/blockquote>\n

\u8ba8\u8bba<\/strong><\/p>\n

\u603b\u4f53\u800c\u8a00\uff0c\u672c\u6587\u505a\u51fa\u4e86\u51e0\u9879\u91cd\u8981\u8d21\u732e\uff1a<\/p>\n

\u7efc\u5408\u8bc4\u4f30\u57fa\u51c6–MedS-Bench<\/strong><\/p>\n

\u533b\u5b66 LLM \u7684\u5f00\u53d1\u5728\u5f88\u5927\u7a0b\u5ea6\u4e0a\u4f9d\u8d56\u4e8e\u591a\u9009\u9898\u56de\u7b54\uff08MCQA\uff09\u7684\u57fa\u51c6\u6d4b\u8bd5\u3002\u7136\u800c\uff0c\u8fd9\u79cd\u72ed\u9698\u7684\u8bc4\u4f30\u6846\u67b6\u56de\u5ffd\u7565 LLM \u5728\u5404\u79cd\u590d\u6742\u4e34\u5e8a\u573a\u666f\u4e0b\u4e2d\u771f\u5b9e\u7684\u80fd\u529b\u8868\u73b0\u3002<\/p>\n

\u56e0\u6b64\uff0c\u5728\u8fd9\u9879\u5de5\u4f5c\u4e2d\uff0c\u7814\u7a76\u56e2\u961f\u5f15\u5165\u4e86 MedS-Bench\uff0c\u8fd9\u662f\u4e00\u4e2a\u7efc\u5408\u57fa\u51c6\uff0c\u65e8\u5728\u8bc4\u4f30\u95ed\u6e90\u548c\u5f00\u6e90 LLM \u5728\u5404\u79cd\u4e34\u5e8a\u4efb\u52a1\u4e2d\u7684\u6027\u80fd\uff0c\u5305\u62ec\u90a3\u4e9b\u9700\u8981\u4ece\u6a21\u578b\u9884\u8bad\u7ec3\u8bed\u6599\u4e2d\u56de\u5fc6\u4e8b\u5b9e\u6216\u4ece\u7ed9\u5b9a\u4e0a\u4e0b\u6587\u4e2d\u8fdb\u884c\u63a8\u7406\u7684\u4efb\u52a1\u3002<\/p>\n

\u7814\u7a76\u7ed3\u679c\u8868\u660e\uff0c\u867d\u7136\u73b0\u6709\u7684 LLM \u5728 MCQA \u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8868\u73b0\u4f18\u5f02\uff0c\u4f46\u5b83\u4eec\u5374\u5f88\u96be\u4e0e\u4e34\u5e8a\u5b9e\u8df5\u4fdd\u6301\u4e00\u81f4\uff0c\u5c24\u5176\u662f\u5728\u6cbb\u7597\u63a8\u8350\u548c\u89e3\u91ca\u7b49\u4efb\u52a1\u4e2d\u3002\u8fd9\u4e00\u53d1\u73b0\u51f8\u663e\u4e86\u8fdb\u4e00\u6b65\u5f00\u53d1\u9002\u914d\u4e8e\u66f4\u5e7f\u6cdb\u4e34\u5e8a\u548c\u533b\u5b66\u573a\u666f\u7684\u533b\u5b66\u5927\u8bed\u8a00\u6a21\u578b\u7684\u5fc5\u8981\u6027\u3002<\/p>\n

\u7efc\u5408\u6307\u4ee4\u8c03\u6574\u6570\u636e\u96c6–MedS-Ins<\/strong><\/p>\n

\u7814\u7a76\u56e2\u961f\u4ece\u73b0\u6709\u7684 BioNLP \u6570\u636e\u96c6\u4e2d\u5e7f\u6cdb\u83b7\u53d6\u6570\u636e\uff0c\u5e76\u5c06\u8fd9\u4e9b\u6837\u672c\u8f6c\u6362\u4e3a\u7edf\u4e00\u683c\u5f0f\uff0c\u540c\u65f6\u91c7\u7528\u534a\u81ea\u52a8\u5316\u7684\u63d0\u793a\u7b56\u7565\uff0c\u6784\u5efa\u5f00\u53d1\u4e86 MedS-Ins–\u4e00\u79cd\u65b0\u578b\u7684\u533b\u7597\u6307\u4ee4\u8c03\u6574\u6570\u636e\u96c6\u3002\u4ee5\u5f80\u7684\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u96c6\u7684\u5de5\u4f5c\u4e3b\u8981\u96c6\u4e2d\u5728\u4ece\u65e5\u5e38\u5bf9\u8bdd\u3001\u8003\u8bd5\u6216\u5b66\u672f\u8bba\u6587\u4e2d\u6784\u5efa\u95ee\u7b54\u5bf9\uff0c\u5f80\u5f80\u5ffd\u7565\u4e86\u4ece\u5b9e\u9645\u4e34\u5e8a\u5b9e\u8df5\u4e2d\u751f\u6210\u7684\u6587\u672c\u3002<\/p>\n

\u76f8\u6bd4\u4e4b\u4e0b\uff0cMedS-Ins \u6574\u5408\u4e86\u66f4\u5e7f\u6cdb\u7684\u533b\u5b66\u6587\u672c\u8d44\u6e90\uff0c\u5305\u62ec 5 \u4e2a\u4e3b\u8981\u6587\u672c\u9886\u57df\u548c 19 \u4e2a\u4efb\u52a1\u7c7b\u522b\u3002\u8fd9\u79cd\u5bf9\u6570\u636e\u7ec4\u6210\u7684\u7cfb\u7edf\u6027\u5206\u6790\u6709\u5229\u4e8e\u7528\u6237\u7406\u89e3 LLM \u7684\u4e34\u5e8a\u5e94\u7528\u8fb9\u754c\u3002<\/p>\n

\u533b\u5b66\u5927\u8bed\u8a00\u6a21\u578b–MMedIns-Llama 3<\/strong><\/p>\n

\u5728\u6a21\u578b\u65b9\u9762\uff0c\u7814\u7a76\u56e2\u961f\u8bc1\u660e\u4e86\u901a\u8fc7\u5728MedS-Ins\u4e0a\u8fdb\u884c\u6307\u4ee4\u5fae\u8c03\u8bad\u7ec3\uff0c\u53ef\u4ee5\u663e\u8457\u63d0\u9ad8\u5f00\u6e90\u533b\u5b66LLM\u4e0e\u4e34\u5e8a\u9700\u6c42\u7684\u4e00\u81f4\u6027\u3002<\/p>\n

\u9700\u8981\u5f3a\u8c03\u7684\u662f\uff0c\u6700\u7ec8\u6a21\u578bMMedIns-Llama 3\u66f4\u591a\u662f\u4e00\u4e2a\u300c\u6982\u5ff5\u9a8c\u8bc1\u300d\u6a21\u578b\uff0c\u5b83\u91c7\u7528\u4e868B\u7684\u4e2d\u7b49\u53c2\u6570\u89c4\u6a21\uff0c\u6700\u7ec8\u7684\u6a21\u578b\u5bf9\u5404\u79cd\u4e34\u5e8a\u4efb\u52a1\u5c55\u73b0\u51fa\u4e86\u6df1\u523b\u7684\u7406\u89e3\uff0c\u5e76\u80fd\u901a\u8fc7\u96f6\u6b21\u6216\u5c11\u91cf\u7684\u6307\u4ee4\u63d0\u793a\u7075\u6d3b\u9002\u5e94\u591a\u79cd\u533b\u7597\u573a\u666f\uff0c\u800c\u65e0\u9700\u8fdb\u4e00\u6b65\u7684\u7279\u5b9a\u4efb\u52a1\u8bad\u7ec3\u3002<\/p>\n

\u7ed3\u679c\u8868\u660e\uff0cMMedIns-Llama 3\u5728\u7279\u5b9a\u4e34\u5e8a\u4efb\u52a1\u7c7b\u578b\u4e0a\u4f18\u4e8e\u73b0\u6709\u7684LLM\u3001 \u5305\u62ec GPT-4\u3001Claude-3.5\u7b49\u3002<\/p>\n

\u73b0\u6709\u7684\u5c40\u9650\u6027<\/strong><\/p>\n

\u5728\u6b64\uff0c\u7814\u7a76\u56e2\u961f\u4e5f\u8981\u5f3a\u8c03\u4e86\u672c\u6587\u7684\u5c40\u9650\u6027\u4ee5\u53ca\u672a\u6765\u53ef\u80fd\u7684\u6539\u8fdb\u3002<\/p>\n

\u9996\u5148\uff0cMedS-Bench \u76ee\u524d\u53ea\u6db5\u76d6\u4e86 11 \u9879\u4e34\u5e8a\u4efb\u52a1\u3001 \u8fd9\u5e76\u4e0d\u80fd\u5b8c\u5168\u6db5\u76d6\u6240\u6709\u4e34\u5e8a\u573a\u666f\u7684\u590d\u6742\u6027\u3002\u6b64\u5916\uff0c\u867d\u7136\u8bc4\u4f30\u4e86\u516d\u79cd\u4e3b\u6d41 LLM\uff0c\u4f46\u5206\u6790\u4e2d\u4ecd\u7136\u7f3a\u5c11\u90e8\u5206\u6700\u65b0\u7684 LLM\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e9b\u5c40\u9650\u6027\uff0c\u7814\u7a76\u56e2\u961f\u8ba1\u5212\u5728\u53d1\u8868\u672c\u6587\u7684\u540c\u65f6\u53d1\u5e03\u4e00\u4e2a\u533b\u5b66 LLM \u7684 Leaderboard\uff0c\u65e8\u5728\u9f13\u52b1\u66f4\u591a\u7684\u7814\u7a76\u4eba\u5458\u4e00\u540c\u4e0d\u65ad\u6269\u5c55\u548c\u5b8c\u5584\u533b\u5b66 LLM \u7684\u7efc\u5408\u8bc4\u4f30\u57fa\u51c6\u3002\u901a\u8fc7\u5728\u8bc4\u4f30\u8fc7\u7a0b\u4e2d\u7eb3\u5165\u66f4\u591a\u6765\u81ea\u4e0d\u540c\u6587\u672c\u6e90\u7684\u4efb\u52a1\u7c7b\u522b\uff0c\u5e0c\u671b\u80fd\u66f4\u6df1\u5165\u5730\u4e86\u89e3\u533b\u5b66\u9886\u57df\u4e2d LLMs \u7684\u5f00\u53d1\u60c5\u51b5\u53ca\u4f7f\u7528\u8fb9\u754c\u3002<\/p>\n

\u5176\u6b21\uff0c\u5c3d\u7ba1\u73b0\u5728 MedS-Ins \u5305\u542b\u4e86\u5e7f\u6cdb\u7684\u533b\u7597\u4efb\u52a1\uff0c\u4f46\u5b83\u4ecd\u7136\u4e0d\u5b8c\u6574\uff0c\u8fd8\u662f\u7f3a\u5c11\u67d0\u4e9b\u5b9e\u7528\u7684\u533b\u7597\u573a\u666f\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\uff0c\u7814\u7a76\u56e2\u961f\u5728 GitHub \u4e0a\u5f00\u6e90\u4e86\u6240\u6709\u6536\u96c6\u5230\u7684\u6570\u636e\u548c\u8d44\u6e90\u3002\u7531\u8877\u5e0c\u671b\u66f4\u591a\u7684\u4e34\u5e8a\u533b\u751f\u6216\u8005\u7814\u7a76\u5b66\u8005\u53ef\u4ee5\u4e00\u540c\u7ef4\u62a4\u6269\u5f20\u8fd9\u4e2a\u6307\u4ee4\u8c03\u6574\u6570\u636e\u96c6\uff0c\u7c7b\u4f3c\u4e8e\u901a\u7528\u9886\u57df\u4e2d\u7684 Super-NaturalInstructions\u3002\u7814\u7a76\u56e2\u961f\u5728 GitHub \u9875\u9762\u4e0a\u63d0\u4f9b\u4e86\u8be6\u7ec6\u7684\u4e0a\u4f20\u6307\u5357\uff0c\u540c\u65f6\u5c06\u5728\u8bba\u6587\u7684\u8fed\u4ee3\u66f4\u65b0\u4e2d\u4e66\u9762\u611f\u8c22\u6bcf\u4e00\u4f4d\u53c2\u4e0e\u6570\u636e\u96c6\u66f4\u65b0\u7684\u8d21\u732e\u8005\u3002<\/p>\n

\u7b2c\u4e09\uff0c\u7814\u7a76\u56e2\u961f\u8ba1\u5212\u5728 MedS-Bench \u548c MedS-Ins \u4e2d\u52a0\u5165\u66f4\u591a\u8bed\u8a00\uff0c\u4ee5\u652f\u6301\u5f00\u53d1\u66f4\u5f3a\u5927\u7684\u591a\u8bed\u8a00\u533b\u5b66 LLM\u3002\u76ee\u524d\uff0c\u5c3d\u7ba1\u5728 MedS-Bench \u548c MedS-Ins \u4e2d\u5305\u542b\u4e86\u4e00\u4e9b\u591a\u8bed\u8a00\u4efb\u52a1\uff0c\u4f46\u8fd9\u4e9b\u8d44\u6e90\u4e3b\u8981\u4ee5\u82f1\u8bed\u4e3a\u4e2d\u5fc3\u3002\u5c06\u5176\u6269\u5c55\u5230\u66f4\u5e7f\u6cdb\u7684\u8bed\u8a00\u8303\u56f4\u5c06\u662f\u4e00\u4e2a\u5f88\u6709\u524d\u666f\u7684\u672a\u6765\u65b9\u5411\uff0c\u4ee5\u4fbf\u786e\u4fdd\u533b\u7597\u4eba\u5de5\u667a\u80fd\u7684\u6700\u65b0\u8fdb\u5c55\u80fd\u591f\u516c\u5e73\u5730\u60e0\u53ca\u66f4\u5e7f\u6cdb\u3001\u66f4\u591a\u6837\u7684\u5730\u533a\u3002<\/p>\n

\u6700\u540e\uff0c\u7814\u7a76\u56e2\u961f\u5df2\u5c06\u6240\u6709\u4ee3\u7801\u3001\u6570\u636e\u548c\u8bc4\u4f30\u6d41\u7a0b\u8fdb\u884c\u5f00\u6e90\u3002\u5e0c\u671b\u8fd9\u9879\u5de5\u4f5c\u80fd\u5f15\u5bfc\u533b\u5b66 LLM \u7684\u5f00\u53d1\u66f4\u591a\u7684\u5173\u6ce8\u5230\u5982\u4f55\u5c06\u8fd9\u4e9b\u5f3a\u5927\u7684\u8bed\u8a00\u6a21\u578b\u4e0e\u73b0\u5b9e\u4e16\u754c\u7684\u4e34\u5e8a\u5e94\u7528\u7ed3\u5408\u3002<\/p>\n

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21\u578b\uff0c\u4e0a\u4ea4\u5927\u56e2\u961f\u53d1\u5e03\u5927\u89c4\u6a21\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u5168\u9762\u57fa\u51c6\u6d4b\u8bd5<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21 […]<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[27],"tags":[68,67],"class_list":["post-35387","post","type-post","status-publish","format-standard","hentry","category-news","tag-68","tag-67"],"yoast_head":"\n\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21\u578b\uff0c\u4e0a\u4ea4\u5927\u56e2\u961f\u53d1\u5e03\u5927\u89c4\u6a21\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u5168\u9762\u57fa\u51c6\u6d4b\u8bd5 - \u4e00\u8d77AI\u6280\u672f<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/17aitech.com\/?p=35387\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=35387\",\"url\":\"https:\/\/17aitech.com\/?p=35387\",\"name\":\"\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21\u578b\uff0c\u4e0a\u4ea4\u5927\u56e2\u961f\u53d1\u5e03\u5927\u89c4\u6a21\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u5168\u9762\u57fa\u51c6\u6d4b\u8bd5 - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=35387#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=35387#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-7798d0c24c3c062c48611621b12fb06d.png\",\"datePublished\":\"2024-12-24T02:59:55+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=35387#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=35387\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=35387#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-7798d0c24c3c062c48611621b12fb06d.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-7798d0c24c3c062c48611621b12fb06d.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=35387#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21\u578b\uff0c\u4e0a\u4ea4\u5927\u56e2\u961f\u53d1\u5e03\u5927\u89c4\u6a21\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u5168\u9762\u57fa\u51c6\u6d4b\u8bd5\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/17aitech.com\/#website\",\"url\":\"https:\/\/17aitech.com\/\",\"name\":\"\u4e00\u8d77AI\u6280\u672f\",\"description\":\"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca\",\"alternateName\":\"\u4e00\u8d77AI\u6280\u672f\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/17aitech.com\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\",\"name\":\"AI\u5c0f\u52a9\u624b\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/image\/\",\"url\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"contentUrl\":\"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png\",\"caption\":\"AI\u5c0f\u52a9\u624b\"},\"description\":\"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e\",\"url\":\"https:\/\/17aitech.com\/?page_id=33738&user=3\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21\u578b\uff0c\u4e0a\u4ea4\u5927\u56e2\u961f\u53d1\u5e03\u5927\u89c4\u6a21\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u5168\u9762\u57fa\u51c6\u6d4b\u8bd5 - \u4e00\u8d77AI\u6280\u672f","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/17aitech.com\/?p=35387","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=35387","url":"https:\/\/17aitech.com\/?p=35387","name":"\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21\u578b\uff0c\u4e0a\u4ea4\u5927\u56e2\u961f\u53d1\u5e03\u5927\u89c4\u6a21\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u5168\u9762\u57fa\u51c6\u6d4b\u8bd5 - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=35387#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=35387#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-7798d0c24c3c062c48611621b12fb06d.png","datePublished":"2024-12-24T02:59:55+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=35387#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=35387"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=35387#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-7798d0c24c3c062c48611621b12fb06d.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2024\/09\/frc-7798d0c24c3c062c48611621b12fb06d.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=35387#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u8fc8\u5411\u300c\u591a\u9762\u624b\u300d\u533b\u7597\u5927\u6a21\u578b\uff0c\u4e0a\u4ea4\u5927\u56e2\u961f\u53d1\u5e03\u5927\u89c4\u6a21\u6307\u4ee4\u5fae\u8c03\u6570\u636e\u3001\u5f00\u6e90\u6a21\u578b\u4e0e\u5168\u9762\u57fa\u51c6\u6d4b\u8bd5"}]},{"@type":"WebSite","@id":"https:\/\/17aitech.com\/#website","url":"https:\/\/17aitech.com\/","name":"\u4e00\u8d77AI\u6280\u672f","description":"\u8ba9AI\u77e5\u8bc6\u89e6\u624b\u53ef\u53ca","alternateName":"\u4e00\u8d77AI\u6280\u672f","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/17aitech.com\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d","name":"AI\u5c0f\u52a9\u624b","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/#\/schema\/person\/image\/","url":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","contentUrl":"\/\/17aitech.com\/wp-content\/uploads\/2024\/04\/robot_3.png","caption":"AI\u5c0f\u52a9\u624b"},"description":"\u8fd9\u4e2a\u4eba\u5f88\u61d2\uff0c\u4ec0\u4e48\u90fd\u6ca1\u6709\u7559\u4e0b\uff5e","url":"https:\/\/17aitech.com\/?page_id=33738&user=3"}]}},"_links":{"self":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/35387","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=35387"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/35387\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=35387"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=35387"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=35387"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}