{"id":802585,"date":"2026-06-22T08:41:57","date_gmt":"2026-06-22T00:41:57","guid":{"rendered":"https:\/\/www.alphaseek.cn\/?p=802585"},"modified":"2026-06-22T08:41:57","modified_gmt":"2026-06-22T00:41:57","slug":"%e5%91%8a%e5%88%ab%e5%8d%95%e7%a7%91%e4%b8%93%e5%ae%b6%ef%bc%9a%e9%a6%96%e4%b8%aaagent%e5%85%a8%e9%9d%a2%e8%bf%9b%e5%8c%96%e6%a1%86%e6%9e%b6eevee%e5%8f%91%e5%b8%83","status":"publish","type":"post","link":"https:\/\/www.alphaseek.cn\/?p=802585","title":{"rendered":"\u544a\u522b\u201c\u5355\u79d1\u4e13\u5bb6\u201d\uff1a\u9996\u4e2aAgent\u5168\u9762\u8fdb\u5316\u6846\u67b6EEVEE\u53d1\u5e03"},"content":{"rendered":"<p><span>\u8fc7\u53bb\u4e24\u5e74\uff0cAI Agent \u7684\u80fd\u529b\u8fb9\u754c\u88ab\u4e0d\u65ad\u5237\u65b0\uff1a\u4f1a\u5199\u4ee3\u7801\u3001\u4f1a\u8c03\u7528\u5de5\u5177\u3001\u4f1a\u53cd\u601d\u5931\u8d25\uff0c\u4e5f\u5f00\u59cb\u80fd\u5728\u4efb\u52a1\u6267\u884c\u4e2d\u79ef\u7d2f\u7ecf\u9a8c\u3002<\/span><\/p>\n<p>\u4f46\u4e00\u4e2a\u66f4\u73b0\u5b9e\u7684\u95ee\u9898\u6b63\u5728\u6d6e\u73b0\uff1a<\/p>\n<p>\u5982\u679c\u4e00\u4e2a Agent \u771f\u7684\u88ab\u90e8\u7f72\u5230\u771f\u5b9e\u4e16\u754c\uff0c\u5b83\u8fd8\u80fd\u7ee7\u7eed\u53d8\u5f3a\u5417\uff1f<\/p>\n<p>\u4e0d\u662f\u5728\u4e00\u4e2a\u56fa\u5b9a benchmark \u4e0a\u53cd\u590d\u5237\u5206\uff0c\u4e5f\u4e0d\u662f\u53ea\u9488\u5bf9\u67d0\u4e00\u79cd\u4efb\u52a1\u6539 prompt\uff0c\u800c\u662f\u5728\u771f\u5b9e\u4f7f\u7528\u4e2d\u540c\u65f6\u9762\u5bf9\u4ee3\u7801\u3001\u6570\u5b66\u3001\u77e5\u8bc6\u95ee\u7b54\u3001\u516c\u5f0f\u8ba1\u7b97\u3001\u590d\u6742\u63a8\u7406\u7b49\u4e0d\u65ad\u53d8\u5316\u7684\u4efb\u52a1\u8f93\u5165\uff0c\u8fd8\u80fd\u4e0d\u80fd\u6301\u7eed\u9002\u5e94\u3001\u7a33\u5b9a\u63d0\u5347\uff1f<\/p>\n<p>\u8fd9\u6b63\u662f EEVEE \u60f3\u89e3\u51b3\u7684\u95ee\u9898\u3002<\/p>\n<p>\u6765\u81ea\u4e0a\u6d77\u4ea4\u901a\u5927\u5b66\u4e0e\u666e\u6797\u65af\u987f\u5927\u5b66\u7684\u7814\u7a76\u56e2\u961f\u53d1\u5e03\u4e86 EEVEE\uff0c\u4e00\u4e2a\u9762\u5411 LLM Agent \u7684\u6d4b\u8bd5\u65f6\u63d0\u793a\u5b66\u4e60\u6846\u67b6\u3002\u5b83\u8bd5\u56fe\u628a prompt learning \u4ece \u201c\u5355\u4e00\u4efb\u52a1\u4f18\u5316\u201d \u63a8\u5411\u66f4\u63a5\u8fd1\u771f\u5b9e\u90e8\u7f72\u7684\u573a\u666f\uff1a\u8ba9\u667a\u80fd\u4f53\u5728\u591a\u7c7b\u578b\u4efb\u52a1\u4e0d\u65ad\u6d8c\u5165\u65f6\uff0c\u4ecd\u7136\u80fd\u591f\u7ee7\u7eed\u5b66\u4e60\uff0c\u800c\u4e0d\u662f\u987e\u6b64\u5931\u5f7c\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/79664B1BC2966DC66541F6F4A66C52FD2B4C336F_size478_w1080_h459.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"\u544a\u522b\u201c\u5355\u79d1\u4e13\u5bb6\u201d\uff1a\u9996\u4e2aAgent\u5168\u9762\u8fdb\u5316\u6846\u67b6EEVEE\u53d1\u5e03\"  \/><\/p>\n<p>\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/2606.11182<\/p>\n<p>\u9879\u76ee\u4e3b\u9875\uff1ahttps:\/\/princeton-ai2-lab.github.io\/EEVEE\/<\/p>\n<p>\u5f00\u6e90\u4ee3\u7801\uff1ahttps:\/\/github.com\/Princeton-AI2-Lab\/EEVEE<\/p>\n<p><strong>\u5355\u4e00\u4efb\u52a1\u4e0a\u7684\u8fdb\u6b65\uff0c\u8fd8\u4e0d\u591f<\/strong><\/p>\n<p>\u4eca\u5929\u5df2\u7ecf\u6709\u5f88\u591a prompt optimization \u65b9\u6cd5\u53ef\u4ee5\u8ba9\u6a21\u578b\u5728\u4e00\u4e2a\u4efb\u52a1\u4e0a\u53d8\u5f97\u66f4\u597d\u3002<\/p>\n<p>\u4f8b\u5982\uff0c\u5728\u4e00\u4e2a\u6570\u5b66\u9898\u96c6\u4e0a\u5b66\u4f1a\u66f4\u4e25\u8c28\u7684\u89e3\u9898\u6b65\u9aa4\uff1b\u5728\u4e00\u4e2a\u4ee3\u7801\u4efb\u52a1\u4e0a\u5b66\u4f1a\u8f93\u51fa\u66f4\u89c4\u8303\u7684\u51fd\u6570\u4f53\uff1b\u5728\u4e00\u4e2a\u95ee\u7b54\u4efb\u52a1\u4e0a\u5b66\u4f1a\u66f4\u7b26\u5408\u8bc4\u6d4b\u683c\u5f0f\u7684\u56de\u7b54\u3002<\/p>\n<p>\u8fd9\u5f88\u6709\u4ef7\u503c\uff0c\u4f46\u5b83\u79bb\u771f\u5b9e Agent \u8fd8\u5dee\u4e00\u6b65\u3002<\/p>\n<p>\u771f\u5b9e\u90e8\u7f72\u4e2d\u7684 Agent \u4e0d\u4f1a\u53ea\u9047\u5230\u4e00\u79cd\u4efb\u52a1\u3002\u5b83\u53ef\u80fd\u4e0a\u4e00\u79d2\u5728\u5199\u4ee3\u7801\uff0c\u4e0b\u4e00\u79d2\u5728\u505a\u91d1\u878d\u516c\u5f0f\u8ba1\u7b97\uff0c\u518d\u4e0b\u4e00\u6b65\u53c8\u8981\u56de\u7b54\u79d1\u5b66\u95ee\u9898\u3002\u4e0d\u540c\u4efb\u52a1\u9700\u8981\u7684\u884c\u4e3a\u5b8c\u5168\u4e0d\u540c\uff1a\u6709\u7684\u8981\u6c42\u4e25\u683c\u8f93\u51fa\u683c\u5f0f\uff0c\u6709\u7684\u8981\u6c42\u7b26\u53f7\u63a8\u7406\uff0c\u6709\u7684\u8981\u6c42\u77e5\u8bc6\u5224\u65ad\uff0c\u6709\u7684\u8981\u6c42\u53ef\u6267\u884c\u4ee3\u7801\u3002<\/p>\n<p>\u5982\u679c\u6240\u6709\u53cd\u9988\u90fd\u88ab\u585e\u8fdb\u540c\u4e00\u4e2a prompt\uff0c\u95ee\u9898\u5c31\u4f1a\u51fa\u73b0\uff1a<\/p>\n<p>\u4e00\u4e2a\u4efb\u52a1\u4e0a\u5b66\u5230\u7684\u7ecf\u9a8c\uff0c\u53ef\u80fd\u4f1a\u4f24\u5bb3\u53e6\u4e00\u4e2a\u4efb\u52a1\u3002<\/p>\n<p>\u6bd4\u5982\uff0c\u4e3a\u4e86\u516c\u5f0f\u9898\u5b66\u5230 \u201c\u53ea\u8f93\u51fa\u6570\u5b57\u201d\uff0c\u53ef\u80fd\u4f1a\u5f71\u54cd\u9700\u8981\u89e3\u91ca\u63a8\u7406\u8fc7\u7a0b\u7684\u9898\uff1b\u4e3a\u4e86\u4ee3\u7801\u9898\u5b66\u5230 \u201c\u53ea\u8f93\u51fa\u51fd\u6570\u4f53\u201d\uff0c\u4e5f\u53ef\u80fd\u4e0d\u9002\u5408\u77e5\u8bc6\u95ee\u7b54\u3002\u4efb\u52a1\u8d8a\u591a\uff0c\u5355\u4e00 prompt \u8d8a\u5bb9\u6613\u53d8\u6210\u4e00\u4e2a\u4e92\u76f8\u51b2\u7a81\u7684\u6742\u7cc5\u4f53\u3002<\/p>\n<p>\u8fd9\u5c31\u662f EEVEE \u8981\u9762\u5bf9\u7684\u6838\u5fc3\u6311\u6218\uff1a<strong>\u667a\u80fd\u4f53\u4e0d\u80fd\u53ea\u5728\u5355\u79d1\u53d8\u5f3a\uff0c\u800c\u8981\u5728\u591a\u79cd\u4efb\u52a1\u4e2d\u4e00\u8d77\u8fdb\u5316\u3002<\/strong><\/p>\n<p><strong>\u4efb\u52a1\u8d8a\u591a\uff0c\u5dee\u8ddd\u8d8a\u660e\u663e<\/strong><\/p>\n<p>EEVEE \u6700\u6709\u8bf4\u670d\u529b\u7684\u7ed3\u679c\uff0c\u4e0d\u662f\u5355\u4e2a\u4efb\u52a1\u4e0a\u7684\u5206\u6570\uff0c\u800c\u662f\u4efb\u52a1\u4e0d\u65ad\u589e\u52a0\u65f6\u7684\u8868\u73b0\u3002<\/p>\n<p>\u7814\u7a76\u56e2\u961f\u628a\u591a\u4e2a\u4e0d\u540c\u7c7b\u578b\u7684\u4efb\u52a1\u4f9d\u6b21\u52a0\u5165\u540c\u4e00\u4e2a\u5b66\u4e60\u8fc7\u7a0b\uff0c\u89c2\u5bdf prompt learning \u65b9\u6cd5\u80fd\u4e0d\u80fd\u6301\u7eed\u7d2f\u79ef\u6536\u76ca\u3002<\/p>\n<p>\u7ed3\u679c\u975e\u5e38\u76f4\u63a5\uff1a<\/p>\n<p>\u5f53\u4efb\u52a1\u8d8a\u6765\u8d8a\u591a\u65f6\uff0c\u4e00\u4e9b\u5f3a\u57fa\u7ebf\u65b9\u6cd5\u7684\u6536\u76ca\u5f00\u59cb\u505c\u6b62\u53e0\u52a0\uff0c\u751a\u81f3\u8dcc\u5230\u8d1f\u6570\uff1b\u800c EEVEE \u4ecd\u7136\u4fdd\u6301\u6b63\u5411\u63d0\u5347\uff0c\u6700\u7ec8\u5728\u4efb\u52a1\u5168\u90e8\u52a0\u5165\u540e\u8fbe\u5230\u7ea6 +42 \u7684\u7d2f\u8ba1\u63d0\u5347\u3002<\/p>\n<p>\u6362\u53e5\u8bdd\u8bf4\uff0cEEVEE \u7684\u4f18\u52bf\u4e0d\u662f \u201c\u5355\u70b9\u63d0\u5206\u201d\uff0c\u800c\u662f\u4efb\u52a1\u53d8\u590d\u6742\u4e4b\u540e\u8fd8\u80fd\u7ee7\u7eed\u5f80\u4e0a\u8d70\u3002<\/p>\n<p>\u8fd9\u975e\u5e38\u5173\u952e\u3002\u56e0\u4e3a\u771f\u5b9e\u4e16\u754c\u91cc\u7684 Agent\uff0c\u9762\u5bf9\u7684\u6c38\u8fdc\u4e0d\u662f\u4e00\u4e2a\u5e72\u51c0\u3001\u5c01\u95ed\u3001\u5355\u4e00\u7684\u4efb\u52a1\u73af\u5883\uff0c\u800c\u662f\u4e0d\u65ad\u53d8\u5316\u7684\u4efb\u52a1\u7ec4\u5408\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/3A2B1BDE8F0D7905C53B02479C810127F4DC83DD_size74_w1080_h349.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"\u544a\u522b\u201c\u5355\u79d1\u4e13\u5bb6\u201d\uff1a\u9996\u4e2aAgent\u5168\u9762\u8fdb\u5316\u6846\u67b6EEVEE\u53d1\u5e03\"  \/><\/p>\n<p>\u5168\u65b9\u4f4d\u63d0\u5347\uff1a\u4e0d\u540c\u6a21\u578b\u4e0a\u90fd\u6709\u6548<\/p>\n<p>\u5728\u4e3b\u5b9e\u9a8c\u4e2d\uff0c\u7814\u7a76\u56e2\u961f\u4f7f\u7528\u4e86\u56db\u7c7b\u4ee3\u8868\u6027\u4efb\u52a1\uff1a\u77e5\u8bc6\u95ee\u7b54\u3001\u516c\u5f0f\u8ba1\u7b97\u3001\u6570\u5b66 \/ \u7b26\u53f7\u63a8\u7406\u3001\u4ee3\u7801\u751f\u6210\u3002\u8fd9\u4e9b\u4efb\u52a1\u653e\u5728\u4e00\u8d77\uff0c\u6784\u6210\u4e86\u4e00\u4e2a\u66f4\u63a5\u8fd1\u771f\u5b9e Agent \u5de5\u4f5c\u8d1f\u8f7d\u7684\u6df7\u5408\u73af\u5883\u3002<\/p>\n<p>\u5728\u8fd9\u4e2a\u8bbe\u7f6e\u4e0b\uff0cEEVEE \u5728\u4e0d\u540c backbone \u4e0a\u90fd\u5e26\u6765\u4e86\u660e\u663e\u63d0\u5347\uff1a<\/p>\n<p>\u5728 Qwen3-4B-Instruct \u4e0a\uff0c\u5e73\u5747\u5206\u4ece 41.37 \u63d0\u5347\u5230 51.75\uff0c\u76f8\u5bf9\u63d0\u5347\u7ea6 25%<\/p>\n<p>\u5728 DeepSeek-V3.2 \u4e0a\uff0c\u5e73\u5747\u5206\u4ece 39.75 \u63d0\u5347\u5230 64.07\uff0c\u76f8\u5bf9\u63d0\u5347\u7ea6 61%<\/p>\n<p>\u76f8\u6bd4\u73b0\u6709 SOTA prompt-learning \u65b9\u6cd5\uff0c\u6700\u9ad8\u76f8\u5bf9\u63d0\u5347\u8fbe\u5230 48.2%<\/p>\n<p>\u8fd9\u8bf4\u660e EEVEE \u5e76\u4e0d\u662f\u53ea\u5bf9\u67d0\u4e2a\u6a21\u578b\u3001\u67d0\u4e2a\u4efb\u52a1\u6709\u6548\uff0c\u800c\u662f\u5728\u66f4\u5e7f\u6cdb\u7684\u6df7\u5408\u4efb\u52a1\u8bbe\u7f6e\u4e2d\uff0c\u90fd\u80fd\u8ba9\u667a\u80fd\u4f53\u83b7\u5f97\u66f4\u7a33\u5b9a\u7684\u6574\u4f53\u63d0\u5347\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/56600BA9746831C610ADA501E1F40B550F5F9943_size162_w1080_h325.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"\u544a\u522b\u201c\u5355\u79d1\u4e13\u5bb6\u201d\uff1a\u9996\u4e2aAgent\u5168\u9762\u8fdb\u5316\u6846\u67b6EEVEE\u53d1\u5e03\"  \/><\/p>\n<p>\u4e0d\u53ea\u662f \u201c\u591a\u5f00\u51e0\u4e2a prompt\u201d\uff1aEEVEE \u5982\u4f55\u8ba9\u667a\u80fd\u4f53\u5206\u5de5\u8fdb\u5316\uff1f<\/p>\n<p>EEVEE \u7684\u601d\u8def\u53ef\u4ee5\u7528\u4e00\u53e5\u8bdd\u6982\u62ec\uff1a<\/p>\n<p><strong>\u5148\u5206\u6d41\uff0c\u518d\u7528\u4e13\u95e8 prompt \u63a8\u7406\u3002<\/strong><\/p>\n<p>\u5b83\u4e0d\u518d\u8ba9\u6240\u6709\u4efb\u52a1\u5171\u4eab\u4e00\u4e2a prompt\uff0c\u800c\u662f\u7ef4\u62a4\u591a\u4e2a specialized prompts\u3002\u6bcf\u5f53\u8f93\u5165\u5230\u6765\uff0cEEVEE \u4f1a\u5148\u5224\u65ad\u5b83\u66f4\u9002\u5408\u54ea\u4e00\u7c7b prompt\uff0c\u518d\u8ba9\u6a21\u578b\u5e26\u7740\u5bf9\u5e94 prompt \u53bb\u56de\u7b54\u3002<\/p>\n<p>\u4f46\u8fd9\u4e0d\u662f\u7b80\u5355\u7684\u4eba\u5de5\u5206\u7c7b\u3002<\/p>\n<p>EEVEE \u4e0d\u4f1a\u4e8b\u5148\u89c4\u5b9a \u201c\u6570\u5b66\u9898\u8d70\u6570\u5b66 prompt\uff0c\u4ee3\u7801\u9898\u8d70\u4ee3\u7801 prompt\u201d\u3002\u56e0\u4e3a\u771f\u5b9e\u4efb\u52a1\u7684\u5dee\u5f02\u5f80\u5f80\u4e0d\u5728\u8868\u9762\u4efb\u52a1\u540d\u79f0\u4e0a\uff0c\u800c\u5728\u4e0d\u540c prompt \u7684\u5b9e\u9645\u8868\u73b0\u4e0a\u3002<\/p>\n<p>\u540c\u6837\u662f\u6570\u5b66\u9898\uff0c\u6709\u7684\u9700\u8981\u516c\u5f0f\u8ba1\u7b97\uff0c\u6709\u7684\u9700\u8981\u5b9a\u7406\u63a8\u7406\uff0c\u6709\u7684\u9700\u8981\u4e25\u683c\u8f93\u51fa\u683c\u5f0f\u3002\u771f\u6b63\u6709\u7528\u7684\u5212\u5206\uff0c\u662f\u5728\u5b66\u4e60\u8fc7\u7a0b\u4e2d\u9010\u6e10\u5f62\u6210\u7684\u3002<\/p>\n<p>\u8fd9\u4e5f\u5e26\u6765\u4e86\u5173\u952e\u96be\u70b9\uff1a<strong>router \u548c prompt \u4e0d\u80fd\u5206\u5f00\u5b66\u3002<\/strong><\/p>\n<p>router \u51b3\u5b9a\u6bcf\u4e2a prompt \u4f1a\u770b\u5230\u54ea\u4e9b\u6837\u672c\uff1bprompt \u7684\u80fd\u529b\u53c8\u51b3\u5b9a\u4e86\u600e\u6837\u7684\u8def\u7531\u662f\u6709\u610f\u4e49\u7684\u3002<\/p>\n<p>\u56e0\u6b64\uff0cEEVEE \u91c7\u7528 router\u2013prompt co-evolution\uff1a\u5148\u4f18\u5316 router\uff0c\u91cd\u65b0\u5212\u5206\u4efb\u52a1\uff1b\u518d\u4f18\u5316\u6bcf\u4e2a prompt\uff1b\u968f\u540e\u628a\u66f4\u65b0\u540e\u7684 prompt \u53cd\u9988\u7ed9\u4e0b\u4e00\u8f6e router\u3002\u8fd9\u4e2a\u8fc7\u7a0b\u4e0d\u65ad\u5faa\u73af\uff0c\u8ba9\u8def\u7531\u66f4\u6e05\u6670\uff0cprompt \u66f4\u4e13\u95e8\u5316\u3002<\/p>\n<p>\u6700\u7ec8\uff0c\u667a\u80fd\u4f53\u4e0d\u662f\u9760\u4e00\u4e2a\u8d8a\u6765\u8d8a\u957f\u3001\u8d8a\u6765\u8d8a\u6df7\u4e71\u7684 prompt \u9002\u5e94\u6240\u6709\u4efb\u52a1\uff0c\u800c\u662f\u628a\u7ecf\u9a8c\u7ec4\u7ec7\u8d77\u6765\uff0c\u8ba9\u4e0d\u540c\u4efb\u52a1\u6a21\u5f0f\u8fdb\u5165\u66f4\u5408\u9002\u7684\u5b66\u4e60\u8def\u5f84\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/F00833485E3E7BA7FCB7EA4909F15F7DB6ACF681_size130_w1080_h417.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"\u544a\u522b\u201c\u5355\u79d1\u4e13\u5bb6\u201d\uff1a\u9996\u4e2aAgent\u5168\u9762\u8fdb\u5316\u6846\u67b6EEVEE\u53d1\u5e03\"  \/><\/p>\n<p>\u5355\u4e00\u4efb\u52a1\u4e0a\u4f9d\u65e7\u4f18\u79c0<\/p>\n<p>\u4e00\u4e2a\u81ea\u7136\u7684\u95ee\u9898\u662f\uff1a\u5982\u679c EEVEE \u662f\u4e3a\u591a\u4efb\u52a1\u8bbe\u8ba1\u7684\uff0c\u5b83\u4f1a\u4e0d\u4f1a\u727a\u7272\u5355\u4efb\u52a1\u80fd\u529b\uff1f<\/p>\n<p>\u5b9e\u9a8c\u663e\u793a\u5e76\u6ca1\u6709\u3002<\/p>\n<p>\u5f53\u6bcf\u4e2a benchmark \u5355\u72ec\u8fdb\u884c prompt learning \u65f6\uff0cEEVEE \u4ecd\u7136\u4fdd\u6301\u5f88\u5f3a\u7684\u7ade\u4e89\u529b\u3002\u4f8b\u5982\uff1a<\/p>\n<p>Formula \u4efb\u52a1\u8fbe\u5230 55.25<\/p>\n<p>HumanEval \u4efb\u52a1\u8fbe\u5230 73.17<\/p>\n<p>TheoremQA \u4ece 14.73 \u63d0\u5347\u5230 25.27<\/p>\n<p>\u8fd9\u8bf4\u660e EEVEE \u4e0d\u662f\u9760 \u201c\u8def\u7531\u201d \u63a9\u76d6 prompt learning \u672c\u8eab\u7684\u4e0d\u8db3\u3002\u76f8\u53cd\uff0c\u5b83\u7684 prompt \u5b66\u4e60\u673a\u5236\u5728\u5355\u4efb\u52a1\u4e0a\u540c\u6837\u6709\u6548\uff1b\u5f53\u4efb\u52a1\u53d8\u591a\u65f6\uff0crouter \u624d\u8fdb\u4e00\u6b65\u53d1\u6325\u7ec4\u7ec7\u7ecf\u9a8c\u3001\u907f\u514d\u4e92\u76f8\u5e72\u6270\u7684\u4f5c\u7528\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/236AEBB78E672AB69C020C1D30BA023DDF04EF22_size32_w1080_h310.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"\u544a\u522b\u201c\u5355\u79d1\u4e13\u5bb6\u201d\uff1a\u9996\u4e2aAgent\u5168\u9762\u8fdb\u5316\u6846\u67b6EEVEE\u53d1\u5e03\"  \/><\/p>\n<p>\u4e0d\u9760\u5806\u780c\u4e0a\u4e0b\u6587<\/p>\n<p>\u5f88\u591a\u8ba9 Agent \u53d8\u5f3a\u7684\u65b9\u6cd5\uff0c\u90fd\u6709\u4e00\u4e2a\u5171\u540c\u526f\u4f5c\u7528\uff1a\u4e0a\u4e0b\u6587\u8d8a\u6765\u8d8a\u957f\u3002<\/p>\n<p>\u7cfb\u7edf\u4e0d\u65ad\u628a\u7ecf\u9a8c\u3001\u89c4\u5219\u3001\u6848\u4f8b\u3001playbook \u8ffd\u52a0\u8fdb prompt\uff0c\u77ed\u671f\u770b\u53ef\u80fd\u6709\u6548\uff0c\u4f46\u4efb\u52a1\u4e00\u591a\uff0c\u6210\u672c\u5c31\u4f1a\u8fc5\u901f\u4e0a\u5347\uff0cprompt \u4e5f\u4f1a\u53d8\u5f97\u5197\u957f\u6df7\u4e71\u3002<\/p>\n<p>EEVEE \u6ca1\u6709\u8d70\u8fd9\u6761\u8def\u3002<\/p>\n<p>\u867d\u7136\u5b83\u589e\u52a0\u4e86\u4e00\u4e2a\u8def\u7531\u6b65\u9aa4\uff0c\u4f46\u6574\u4f53 token \u6210\u672c\u4ecd\u7136\u4fdd\u6301\u5728\u8f83\u4f4e\u6c34\u5e73\u3002\u5b9e\u9a8c\u663e\u793a\uff0cEEVEE \u5e73\u5747\u6bcf\u4e2a\u6d4b\u8bd5\u6837\u672c\u4f7f\u7528 <strong>4.32K tokens<\/strong>\uff0c\u63a5\u8fd1\u9ad8\u6548 prompt-learning \u57fa\u7ebf GEPA \u7684 <strong>3.47K<\/strong>\uff0c\u8fdc\u4f4e\u4e8e ACE \u7684 <strong>21.30K<\/strong>\u3002<\/p>\n<p>\u4e5f\u5c31\u662f\u8bf4\uff0cEEVEE \u7684\u63d0\u5347\u5e76\u4e0d\u662f\u9760\u65e0\u9650\u6269\u5c55\u4e0a\u4e0b\u6587\u5806\u51fa\u6765\u7684\uff0c\u800c\u662f\u6765\u81ea\u66f4\u6709\u6548\u7684\u4efb\u52a1\u7ec4\u7ec7\u548c prompt specialization\u3002<\/p>\n<p><img data-lazyload=\"https:\/\/x0.ifengimg.com\/res\/2026\/06358571346FDB3E4A9A9ED6F13C37F5CE569590_size67_w1080_h440.png\" src=\"https:\/\/tech.ifeng.com\/c\/data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABAQMAAAAl21bKAAAAA1BMVEXy8vJkA4prAAAACklEQVQI12NgAAAAAgAB4iG8MwAAAABJRU5ErkJggg==\" style=\"display:block; margin-left:auto; margin-right:auto\" alt=\"\u544a\u522b\u201c\u5355\u79d1\u4e13\u5bb6\u201d\uff1a\u9996\u4e2aAgent\u5168\u9762\u8fdb\u5316\u6846\u67b6EEVEE\u53d1\u5e03\"  \/><\/p>\n<p>Prompt learning \u771f\u6b63\u5b66\u5230\u7684\u662f\u4ec0\u4e48\uff1f<\/p>\n<p>\u8bba\u6587\u4e2d\u7684\u6848\u4f8b\u5206\u6790\u4e5f\u63ed\u793a\u4e86\u4e00\u4e2a\u5f88\u91cd\u8981\u7684\u73b0\u8c61\uff1a<\/p>\n<p>Prompt learning \u6700\u64c5\u957f\u5b66\u4e60\u7684\uff0c\u4e0d\u662f\u51ed\u7a7a\u8865\u5145\u77e5\u8bc6\uff0c\u800c\u662f\u628a\u53cd\u9988\u8f6c\u5316\u6210\u53ef\u590d\u7528\u7684\u505a\u4e8b\u65b9\u5f0f\u3002<\/p>\n<p>\u5728\u4ee3\u7801\u4efb\u52a1\u4e2d\uff0c\u5b83\u80fd\u5b66\u4f1a\u4fdd\u6301\u51fd\u6570\u63a5\u53e3\u3001\u8f93\u51fa\u53ef\u6267\u884c\u4ee3\u7801\u3001\u5904\u7406\u8fb9\u754c\u6761\u4ef6\uff1b\u5728\u516c\u5f0f\u4efb\u52a1\u4e2d\uff0c\u5b83\u80fd\u5b66\u4f1a\u6b63\u786e\u5957\u516c\u5f0f\u3001\u4fdd\u6301\u5355\u4f4d\u5c3a\u5ea6\u3001\u8f93\u51fa\u7b26\u5408\u8bc4\u6d4b\u8981\u6c42\u7684\u7b54\u6848\u683c\u5f0f\u3002<\/p>\n<p>\u4f46\u5bf9\u4e8e\u77e5\u8bc6\u5bc6\u96c6\u578b\u95ee\u7b54\uff0c\u60c5\u51b5\u4f1a\u66f4\u590d\u6742\u3002Prompt learning \u53ef\u4ee5\u8ba9\u6a21\u578b\u63a8\u7406\u66f4\u7cfb\u7edf\uff0c\u5374\u4e0d\u4e00\u5b9a\u80fd\u8865\u4e0a\u7f3a\u5931\u7684\u9886\u57df\u77e5\u8bc6\u3002\u5982\u679c\u6a21\u578b\u672c\u8eab\u7f3a\u5c11\u5173\u952e\u4e8b\u5b9e\uff0c\u5355\u9760 prompt \u4e0d\u4e00\u5b9a\u80fd\u89e3\u51b3\u3002<\/p>\n<p>\u8fd9\u8bf4\u660e EEVEE \u5e76\u4e0d\u662f\u5728\u5ba3\u79f0 prompt learning \u53ef\u4ee5\u89e3\u51b3\u4e00\u5207\u3002\u5b83\u771f\u6b63\u5c55\u793a\u7684\u662f\uff1a\u5f53\u53cd\u9988\u80fd\u591f\u53d8\u6210\u53ef\u590d\u7528\u7684\u8fc7\u7a0b\u3001\u683c\u5f0f\u548c<a href=\"https:\/\/www.alphaseek.cn\/?tag=%e7%ad%96%e7%95%a5\" title=\"View all posts in \u7b56\u7565\" target=\"_blank\" class=\"tag_link\">\u7b56\u7565<\/a>\u65f6\uff0c\u667a\u80fd\u4f53\u53ef\u4ee5\u5728\u6d4b\u8bd5\u65f6\u6301\u7eed\u5438\u6536\u8fd9\u4e9b\u7ecf\u9a8c\uff0c\u5e76\u628a\u5b83\u4eec\u7ec4\u7ec7\u5230\u66f4\u5408\u9002\u7684 prompt \u4e2d\u3002<\/p>\n<p>\u8fd9\u6bd4\u5355\u7eaf \u201c\u6539\u4e00\u4e2a prompt\u201d \u66f4\u63a5\u8fd1\u771f\u5b9e Agent \u7684\u5b66\u4e60\u65b9\u5f0f\u3002<\/p>\n<\/p>\n<p>\u8ba9 Agent \u4ece\u5355\u79d1\u8fdb\u6b65\u8d70\u5411\u5168\u65b9\u4f4d\u8fdb\u5316<\/p>\n<p>\u8fc7\u53bb\u7684 prompt optimization\uff0c\u66f4\u50cf\u662f\u5728\u8bad\u7ec3\u4e00\u4e2a Agent \u505a\u597d\u67d0\u4e00\u95e8\u8bfe\u3002<\/p>\n<p>\u5b83\u53ef\u4ee5\u5728\u4e00\u4e2a\u56fa\u5b9a\u4efb\u52a1\u4e0a\u53cd\u590d\u7ec3\u4e60\uff0c\u9010\u6e10\u6478\u6e05\u89c4\u5f8b\uff0c\u62ff\u5230\u66f4\u9ad8\u5206\u6570\u3002<\/p>\n<p>\u4f46\u771f\u5b9e\u4e16\u754c\u8981\u6c42\u7684\u4e0d\u662f \u201c\u5355\u79d1\u63d0\u5206\u201d\uff0c\u800c\u662f\u66f4\u63a5\u8fd1 \u201c\u5168\u79d1\u6210\u957f\u201d\uff1a\u4efb\u52a1\u4e0d\u65ad\u53d8\u5316\uff0c\u53cd\u9988\u4e0d\u65ad\u53d8\u5316\uff0c\u80fd\u529b\u9700\u6c42\u4e5f\u4e0d\u65ad\u53d8\u5316\u3002\u4e00\u4e2a\u771f\u6b63\u6709\u7528\u7684 self-improving Agent\uff0c\u5fc5\u987b\u80fd\u5728\u90e8\u7f72\u540e\u7ee7\u7eed\u9002\u5e94\u8fd9\u4e9b\u53d8\u5316\uff0c\u800c\u4e0d\u662f\u53ea\u5728\u5355\u4e00 benchmark \u4e0a\u8d8a\u6765\u8d8a\u719f\u7ec3\u3002<\/p>\n<p>EEVEE \u7684\u610f\u4e49\u5c31\u5728\u4e8e\u6b64\u3002<\/p>\n<p>\u5b83\u628a test-time prompt learning \u4ece\u5355\u4e00\u4efb\u52a1\u63a8\u5411\u591a\u7c7b\u578b\u4efb\u52a1\u5171\u540c\u5230\u6765\u7684\u573a\u666f\uff0c\u8ba9\u667a\u80fd\u4f53\u5f00\u59cb\u5b66\u4e60\u5982\u4f55\u7ec4\u7ec7\u7ecf\u9a8c\u3001\u533a\u5206\u4efb\u52a1\u6a21\u5f0f\u3001\u4fdd\u7559\u4e0d\u540c\u80fd\u529b\uff0c\u5e76\u5728\u590d\u6742\u4efb\u52a1\u6d41\u4e2d\u7ee7\u7eed\u63d0\u5347\u6574\u4f53\u8868\u73b0\u3002<\/p>\n<p>\u8fd9\u8fd8\u4e0d\u662f\u6700\u7ec8\u5f62\u6001\u3002\u8bba\u6587\u4e5f\u6307\u51fa\uff0cEEVEE \u4ecd\u7136\u4f9d\u8d56 ground-truth \u6216 rule-based feedback\uff0c\u8fd8\u4e0d\u662f\u5b8c\u5168\u5728\u7ebf\u3001\u5b8c\u5168\u81ea\u76d1\u7763\u7684\u81ea\u6211\u63d0\u5347\u7cfb\u7edf\u3002<\/p>\n<p>\u4f46\u5b83\u8fc8\u51fa\u4e86\u5173\u952e\u4e00\u6b65\uff1a\u5f53\u667a\u80fd\u4f53\u771f\u6b63\u9762\u5bf9\u771f\u5b9e\u4e16\u754c\u4e2d\u590d\u6742\u3001\u591a\u6837\u3001\u4e0d\u65ad\u53d8\u5316\u7684\u4efb\u52a1\u65f6\uff0cprompt learning \u4ecd\u7136\u53ef\u4ee5\u6210\u4e3a\u4e00\u79cd\u6709\u6548\u7684\u6301\u7eed\u9002\u5e94\u673a\u5236\u3002<\/p>\n<p>\u4ece\u5355\u4e00\u4efb\u52a1\u5230\u5168\u65b9\u4f4d\u8fdb\u5316\uff0c\u8fd9\u6b63\u662f EEVEE \u60f3\u6253\u5f00\u7684\u4e00\u6247\u95e8\u3002<\/p>\n<p>(\u6765\u6e90\uff1a\u65b0\u6d6a\u79d1\u6280)<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u8fc7\u53bb\u4e24\u5e74\uff0cAI Agent \u7684\u80fd\u529b\u8fb9\u754c\u88ab\u4e0d\u65ad\u5237\u65b0\uff1a\u4f1a\u5199\u4ee3\u7801\u3001\u4f1a\u8c03\u7528\u5de5\u5177\u3001\u4f1a\u53cd\u601d\u5931\u8d25\uff0c\u4e5f\u5f00\u59cb\u80fd\u5728\u4efb\u52a1\u6267\u884c\u4e2d\u79ef\u7d2f\u7ecf [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[6],"tags":[88],"views":3,"_links":{"self":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/posts\/802585"}],"collection":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=802585"}],"version-history":[{"count":0,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=\/wp\/v2\/posts\/802585\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=802585"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=802585"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.alphaseek.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=802585"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}