{"id":967,"date":"2025-08-21T09:40:00","date_gmt":"2025-08-21T01:40:00","guid":{"rendered":"http:\/\/www.applicationsofllm.com\/?p=967"},"modified":"2025-09-17T09:43:31","modified_gmt":"2025-09-17T01:43:31","slug":"%e4%bb%8e%e3%80%8c%e4%bc%9a%e8%af%b4%e3%80%8d%e8%bf%88%e5%90%91%e3%80%8c%e4%bc%9a%e5%81%9a%e3%80%8d%ef%bc%8cllm%e4%b8%8b%e5%8d%8a%e5%9c%ba%ef%bc%9aagentic%e5%bc%ba%e5%8c%96%e5%ad%a6%e4%b9%a0%e8%8c%83","status":"publish","type":"post","link":"http:\/\/www.applicationsofllm.com\/index.php\/2025\/08\/21\/%e4%bb%8e%e3%80%8c%e4%bc%9a%e8%af%b4%e3%80%8d%e8%bf%88%e5%90%91%e3%80%8c%e4%bc%9a%e5%81%9a%e3%80%8d%ef%bc%8cllm%e4%b8%8b%e5%8d%8a%e5%9c%ba%ef%bc%9aagentic%e5%bc%ba%e5%8c%96%e5%ad%a6%e4%b9%a0%e8%8c%83\/","title":{"rendered":"\u4ece\u300c\u4f1a\u8bf4\u300d\u8fc8\u5411\u300c\u4f1a\u505a\u300d\uff0cLLM\u4e0b\u534a\u573a\uff1aAgentic\u5f3a\u5316\u5b66\u4e60\u8303\u5f0f\u7efc\u8ff0"},"content":{"rendered":"\n<p>\u8fc7\u53bb\u51e0\u5e74\uff0c\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u8bad\u7ec3\u5927\u591a\u4f9d\u8d56\u4e8e<strong>\u57fa\u4e8e\u4eba\u7c7b\u6216\u6570\u636e\u504f\u597d\u7684\u5f3a\u5316\u5b66\u4e60\uff08Preference-based Reinforcement Fine-tuning, PBRFT\uff09<\/strong>\uff1a\u8f93\u5165\u63d0\u793a\u3001\u8f93\u51fa\u6587\u672c\u3001\u83b7\u5f97\u4e00\u4e2a\u504f\u597d\u5206\u6570\u3002\u8fd9\u4e00\u8303\u5f0f\u50ac\u751f\u4e86 GPT-4\u3001Llama-3 \u7b49\u6210\u529f\u7684\u65e9\u671f\u5927\u6a21\u578b\uff0c\u4f46\u5c40\u9650\u4e5f\u65e5\u76ca\u660e\u663e\uff1a<strong>\u7f3a\u4e4f\u957f\u671f\u89c4\u5212\u3001\u73af\u5883\u4ea4\u4e92\u4e0e\u6301\u7eed\u5b66\u4e60\u80fd\u529b<\/strong>\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u7a81\u7834\u8fd9\u6837\u7684\u74f6\u9888\uff0c\u81ea 2025 \u5e74\u521d DeepSeek R1 \u53ca\u5176\u80cc\u540e\u7684 GRPO \u8303\u5f0f\u83b7\u5f97\u7a7a\u524d\u70ed\u5ea6\u4ee5\u540e\uff0c\u4e00\u79cd\u65b0\u7684\u8bad\u7ec3\u8303\u5f0f \u2014\u2014<strong>Agentic Reinforcement Learning<\/strong><strong>\uff08Agentic RL\uff09<\/strong>\uff0c\u6108\u53d1\u5230\u793e\u533a\u5173\u6ce8\u3002\u5b83\u8bd5\u56fe\u8ba9 LLM \u4ece\u300c\u88ab\u52a8\u5bf9\u9f50\u300d\u8fdb\u5316\u4e3a\u300c\u4e3b\u52a8\u51b3\u7b56\u300d\u7684\u667a\u80fd\u4f53\uff0c\u5728\u52a8\u6001\u73af\u5883\u4e2d\u89c4\u5212\u3001\u884c\u52a8\u3001\u5b66\u4e60\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u8bba\u6587\u6807\u9898\uff1aThe Landscape of Agentic Reinforcement Learning for LLMs: A Survey<\/li>\n\n\n\n<li>arXiv \u5730\u5740\uff1ahttps:\/\/arxiv.org\/pdf\/2509.02547<\/li>\n\n\n\n<li>GitHub \u5730\u5740\uff1ahttps:\/\/github.com\/xhyumiracle\/Awesome-AgenticLLM-RL-Papers<\/li>\n<\/ul>\n\n\n\n<p>\u4e3a\u4e86\u634b\u6e05\u8fd9\u4e00\u65b0\u5174\u9886\u57df\uff0c\u4e00\u7bc7\u957f\u8fbe 100 \u9875\u3001\u7531\u725b\u6d25\u5927\u5b66\u3001\u65b0\u52a0\u5761\u56fd\u7acb\u5927\u5b66\u3001\u4f0a\u5229\u8bfa\u4f0a\u5927\u5b66\u5384\u5df4\u7eb3-\u9999\u69df\u5206\u6821\uff0c\u4f26\u6566\u5927\u5b66\u5b66\u9662\u3001\u5e1d\u56fd\u7406\u5de5\u5b66\u9662\u3001\u4e0a\u6d77\u4eba\u5de5\u667a\u80fd\u5b9e\u9a8c\u5ba4\u7b49 16 \u5bb6\u6d77\u5185\u5916\u9876\u7ea7\u7814\u7a76\u673a\u6784\u8054\u5408\u5b8c\u6210\u7684\u6700\u65b0\u7efc\u8ff0\u8bba\u6587\uff0c\u5168\u9762\u7cfb\u7edf\u5730\u68b3\u7406\u4e86\u4f5c\u7528\u4e8e LLM \u7684 Agentic RL \u8fd9\u4e00\u65b9\u5411\uff0c\u8986\u76d6 500 + \u76f8\u5173\u7814\u7a76\uff0c\u6784\u5efa\u4e86 Agentic RL \u7684\u7406\u8bba\u6846\u67b6\u3001\u6f14\u5316\u8109\u7edc\u4e0e\u8d44\u6e90\u7248\u56fe\uff0c\u5e76\u8ba8\u8bba\u4e86\u53ef\u4fe1\u6027\u3001\u6269\u5c55\u6027\u548c\u590d\u6742\u73af\u5883\u7b49\u672a\u6765\u6311\u6218\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"882\" height=\"1024\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-267-882x1024.jpg\" alt=\"\" class=\"wp-image-969\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-267-882x1024.jpg 882w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-267-258x300.jpg 258w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-267-768x891.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-267.jpg 1108w\" sizes=\"auto, (max-width: 882px) 100vw, 882px\" \/><\/figure>\n\n\n\n<p><strong>\u8303\u5f0f\u8fc1\u79fb\uff1a\u4ece PBRFT \u5230 Agentic RL<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"555\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-270-1024x555.jpg\" alt=\"\" class=\"wp-image-972\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-270-1024x555.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-270-300x163.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-270-768x417.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-270.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><sup>&nbsp; &nbsp; &nbsp; \u4ece LLM-RL \u5230 Agentic RL \u8303\u5f0f\u8fc1\u79fb\u6982\u89c8<\/sup><\/p>\n\n\n\n<p>\u8be5\u7efc\u8ff0\u9996\u5148\u7ed9\u51fa\u8303\u5f0f\u8fc1\u79fb\u7684\u5f62\u5f0f\u5316\u5b9a\u4e49\uff1a\u65e9\u671f RL \u7814\u7a76\u591a\u57fa\u4e8e PBRFT \u8303\u5f0f\uff0c\u53ef\u88ab\u89c6\u4e3a\u9000\u5316\u7684\u5355\u6b65 MDP\uff08\u5355 prompt\u3001\u4e00\u6b21\u6027\u6587\u672c\u8f93\u51fa\u3001\u7acb\u5373\u7ec8\u6b62\uff09\uff0c\u800c Agentic RL \u5219\u5c06 LLM \u7f6e\u4e8e<strong>\u90e8\u5206\u53ef\u89c2\u6d4b\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\uff08POMDP\uff09\u4e0b\u8fdb\u884c\u591a\u6b65\u4ea4\u4e92<\/strong>\uff1a<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"196\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-268-1024x196.jpg\" alt=\"\" class=\"wp-image-970\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-268-1024x196.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-268-300x57.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-268-768x147.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-268.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p>\u5176\u4e2d\u5173\u952e\u53d8\u5316\u5728\u4e8e<strong>\u52a8\u4f5c<\/strong>\u7a7a\u95f4\u4ece\u5355\u4e00\u6587\u672c\u6269\u5c55\u4e3a\u300c\u6587\u672c + \u64cd\u4f5c\u300d\uff08<img loading=\"lazy\" decoding=\"async\" width=\"554\" height=\"96\" src=\"blob:http:\/\/www.applicationsofllm.com\/1f42142d-ce0f-4f06-9061-8a62dc35f173\" alt=\"\u56fe\u7247\">\uff09\uff1b\u540c\u65f6<strong>\u5956\u52b1<\/strong>\u4ece\u300c\u5355\u6b65\u8bc4\u5206\u300d\u6269\u5c55\u4e3a\u300c\u65f6\u5e8f\u53cd\u9988\u300d\uff0c\u4f18\u5316<strong>\u6574\u6761\u51b3\u7b56\u8f68\u8ff9<\/strong>\uff0c\u628a LLM \u4ece\u300c\u6587\u672c\u751f\u6210\u5668\u300d\u63a8\u8fdb\u4e3a\u300c\u53ef\u4ea4\u4e92\u7684\u51b3\u7b56\u4f53\u300d\u3002<\/p>\n\n\n\n<p>\u4e00\u53e5\u8bdd\uff1a<strong>PBRFT \u8ba9\u6a21\u578b\u66f4\u4f1a\u4e00\u6b21\u5730\u8bf4\uff0cAgentic RL \u8ba9\u6a21\u578b\u66f4\u4f1a\u957f\u7a0b\u5730\u505a<\/strong>\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"299\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-269-1024x299.jpg\" alt=\"\" class=\"wp-image-971\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-269-1024x299.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-269-300x87.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-269-768x224.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-269.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p>\u5728<strong>\u5f3a\u5316\u5b66\u4e60\u4f18\u5316\u7b97\u6cd5<\/strong>\u5c42\u9762\uff0c\u5f53\u524d\u5b9e\u8df5\u5f62\u6210\u4e86\u4e00\u6761\u4ece\u901a\u7528\u7b56\u7565\u68af\u5ea6\u5230\u504f\u597d\u4f18\u5316\u7684\u8c31\u7cfb\uff0cTable 2 \u6c47\u603b\u6bd4\u8f83\u4e86\u4e09\u7c7b\u7b97\u6cd5\u5bb6\u65cf\u53ca\u5176\u4ee3\u8868\u65b9\u6cd5\uff0c\u4fbf\u4e8e\u8bfb\u8005\u5feb\u901f\u5bf9\u7167\u300c\u8bad\u7ec3\u7528\u4ec0\u4e48\u7b97\u6cd5\u300d\u4e0e\u300c\u5bf9\u9f50\u76ee\u6807\/\u4fe1\u53f7\u5f62\u6001\u300d \u7684\u5bf9\u5e94\u5173\u7cfb\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"1000\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-271-1024x1000.jpg\" alt=\"\" class=\"wp-image-973\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-271-1024x1000.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-271-300x293.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-271-768x750.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-271.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><strong>\u516d\u5927\u6838\u5fc3\u80fd\u529b\uff1a\u667a\u80fd\u4f53\u7684\u300c\u5185\u529f\u300d<\/strong><\/p>\n\n\n\n<p>\u8981\u8ba9 LLM \u771f\u6b63\u6210\u4e3a\u667a\u80fd\u4f53\uff0c\u4ec5\u6709\u52a8\u4f5c\u7a7a\u95f4\u8fd8\u4e0d\u591f\uff0c\u5b83\u5fc5\u987b\u53d1\u5c55\u51fa\u4e00\u5957\u5b8c\u6574\u7684\u80fd\u529b\u4f53\u7cfb\u3002\u8be5\u7efc\u8ff0\u5c06\u5176\u603b\u7ed3\u4e3a\u516d\u5927\u6838\u5fc3\u6a21\u5757\uff0c\u5e76\u5bf9\u6bcf\u4e2a\u6a21\u5757\u63d0\u51fa\u4e86\u524d\u77bb\u6027\u8ba8\u8bba\uff1a<\/p>\n\n\n\n<p><strong>1. \u89c4\u5212\uff08Planning\uff09<\/strong>\uff1a\u4e3a\u590d\u6742\u4efb\u52a1\u8bbe\u5b9a\u5b50\u76ee\u6807\u4e0e\u591a\u6b65\u884c\u52a8\u5e8f\u5217\u3002\u901a\u8fc7<strong>\u5916\u90e8\u5f15\u5bfc<\/strong>\uff08\u5916\u90e8\u6253\u5206\u751f\u6210\u5956\u52b1\uff09\u6216\u5185\u90e8\u9a71\u52a8\uff08\u81ea\u4e3b\u89c4\u5212\u5e76\u4fee\u6b63\uff09\u5b9e\u73b0\u3002<\/p>\n\n\n\n<p><strong>2. \u5de5\u5177\u4f7f\u7528\uff08Tool Use\uff09<\/strong>\uff1a\u8c03\u7528\u5916\u90e8\u5de5\u5177\u5b8c\u6210\u4efb\u52a1\u3002\u4ece ReAct \u7b49\u9759\u6001\u63d0\u793a\u6a21\u4eff\u6f14\u8fdb\u5230 Tool-integrated RL (TIR)\uff0c\u8ba9\u667a\u80fd\u4f53\u5b66\u4f1a\u81ea\u4e3b\u9009\u62e9\u7ec4\u5408\u5de5\u5177\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"272\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-272-1024x272.jpg\" alt=\"\" class=\"wp-image-974\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-272-1024x272.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-272-300x80.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-272-768x204.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-272.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><sup>&nbsp; &nbsp; &nbsp; Agentic Tool Using \u6f14\u5316\u8def\u5f84<\/sup><\/p>\n\n\n\n<p><strong>3. \u8bb0\u5fc6\uff08Memory\uff09<\/strong>\uff1a\u4fdd\u6301\u4e0a\u4e0b\u6587\u8fde\u8d2f\u5e76\u79ef\u7d2f\u77e5\u8bc6\uff0c\u5305\u62ec<strong>\u57fa\u4e8e\u5916\u90e8\u6570\u636e\u5e93\u68c0\u7d22\u8bb0\u5fc6\u3001Token \u7ea7\u522b\u8bb0\u5fc6\u548c\u7ed3\u6784\u5316\u8bb0\u5fc6<\/strong>\u3002\u5176\u4e2d\uff0c\u503c\u5f97\u5173\u6ce8\u7684\u5de5\u4f5c\u5305\u62ec\u6765\u81ea\u5b57\u8282\u8df3\u52a8\u7684 MemAgent \u548c\u9ebb\u7701\u7406\u5de5\u5927\u5b66\u7684 MEM1\uff0c\u4ed6\u4eec\u90fd\u901a\u8fc7\u5f3a\u5316\u5b66\u4e60\u8ba9 LLM Agent \u62e5\u6709\u81ea\u884c\u7ba1\u7406\u8bb0\u5fc6\u7a97\u53e3\u7684\u80fd\u529b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"658\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-273-1024x658.jpg\" alt=\"\" class=\"wp-image-975\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-273-1024x658.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-273-300x193.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-273-768x494.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-273.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><strong>4. \u81ea\u6211\u6539\u8fdb\uff08Self-Improvement\uff09<\/strong>\u540c\u6837\u662f\u76ee\u524d Agent \u6700\u70ed\u95e8\u7684\u53d1\u5c55\u65b9\u5411\u3002\u8be5\u7efc\u8ff0\u9ad8\u5c4b\u5efa\u74f4\u5730\u5c06\u76ee\u524d Agent \u81ea\u6211\u63d0\u5347\u7684\u80fd\u529b\u5212\u5206\u4e3a\u4ee5\u4e0b\u4e09\u7c7b\uff1a<\/p>\n\n\n\n<ol start=\"1\" style=\"list-style-type:lower-roman\" class=\"wp-block-list\">\n<li>\u57fa\u4e8e\u8bed\u8a00\u5f3a\u5316\u5b66\u4e60\uff0c\u5373\u7c7b\u4f3c\u4e8e Reflexion\u3001Self-Critic \u7b49\u98ce\u683c\u7684\u81ea\u6211\u7ea0\u6b63\uff1b<\/li>\n\n\n\n<li>\u901a\u8fc7\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u5185\u5316\u81ea\u63d0\u5347\u80fd\u529b\uff0c\u8b6c\u5982\u6765\u81ea MIT-IBM Watson AI Lab \u7684 Satori \u4fbf\u901a\u8fc7\u5f3a\u5316\u5b66\u4e60\u5185\u5316 Agent \u5728\u6d4b\u8bd5\u9636\u6bb5\u81ea\u6211\u7ea0\u6b63\u7684\u80fd\u529b\uff1b\u7c7b\u4f3c\u7684\u5de5\u4f5c\u8fd8\u6709\u6765\u81ea\u4e0a\u6d77 AI Lab \u7684 TTRL\uff0cMeta \u7684 SWEET-RL \u7b49\u7b49\uff1b<\/li>\n\n\n\n<li>\u901a\u8fc7\u8fed\u4ee3\u81ea\u8bad\u7ec3\uff0c\u8b6c\u5982\u6765\u81ea\u6e05\u534e\u7684 Absolute Zero\u3001\u6765\u81ea\u65af\u5766\u798f\u7684 Sirius \u7b49\u7b49\u3002<\/li>\n<\/ol>\n\n\n\n<p><strong>5. \u63a8\u7406\uff08Reasoning\uff09<\/strong>\uff1a\u89e3\u51b3\u590d\u6742\u95ee\u9898\u7684\u63a8\u5bfc\u80fd\u529b\uff0c\u5206\u4e3a\u5feb\u901f\u76f4\u89c9\u63a8\u7406\uff08\u51ed\u7ecf\u9a8c\u76f4\u89c9\u8fc5\u901f\u7b54\u9898\uff09\u548c\u6162\u901f\u7f1c\u5bc6\u63a8\u7406\uff08\u591a\u6b65\u6f14\u7ece\u5f97\u51fa\u4e25\u8c28\u7ed3\u8bba\uff09\u3002<\/p>\n\n\n\n<p><strong>6. \u611f\u77e5\uff08Perception\uff09<\/strong>\uff1a\u7406\u89e3\u591a\u6a21\u6001\u8f93\u5165\u7684\u4fe1\u606f\u83b7\u53d6\u80fd\u529b\u3002\u6a21\u578b\u4ece\u88ab\u52a8\u8bc6\u522b\u8d70\u5411\u4e3b\u52a8\u611f\u77e5\uff0c\u53ef\u901a\u8fc7<strong>\u5b9a\u4f4d\u9a71\u52a8<\/strong>\uff08\u5c06\u63a8\u7406\u951a\u5b9a\u5177\u4f53\u5bf9\u8c61\uff09\u3001<strong>\u5de5\u5177\u9a71\u52a8<\/strong>\uff08\u501f\u52a9\u5916\u90e8\u5de5\u5177\u8f85\u52a9\uff09\u548c<strong>\u751f\u6210\u9a71\u52a8<\/strong>\uff08\u751f\u6210\u56fe\u50cf\u8349\u56fe\u8f85\u52a9\u63a8\u7406\uff09\u7b49\u65b9\u5f0f\u63d0\u5347\u611f\u77e5\u6548\u679c\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"405\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-274-1024x405.jpg\" alt=\"\" class=\"wp-image-976\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-274-1024x405.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-274-300x119.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-274-768x304.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-274.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><sup>&nbsp; &nbsp; &nbsp; \u667a\u80fd\u4f53\u4e0e\u73af\u5883\u4ea4\u4e92\u95ed\u73af\u793a\u610f<\/sup><\/p>\n\n\n\n<p><strong>\u501f\u52a9\u5f3a\u5316\u5b66\u4e60\uff0c\u8fd9\u4e9b\u80fd\u529b\u7531\u4eba\u5de5\u542f\u53d1\u5f0f\u8f6c\u53d8\u4e3a\u53ef\u5b66\u4e60\u7684\u7b56\u7565<\/strong>\uff0c\u89c4\u5212\u4e0d\u518d\u4f9d\u8d56\u786c\u7f16\u7801\u6d41\u7a0b\u3001\u5de5\u5177\u4f7f\u7528\u4e5f\u53ef\u7531\u6a21\u578b\u81ea\u4e3b\u51b3\u5b9a\u3001\u7aef\u5230\u7aef\u8bad\u7ec3\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1022\" height=\"1024\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-275-1022x1024.jpg\" alt=\"\" class=\"wp-image-977\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-275-1022x1024.jpg 1022w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-275-300x300.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-275-150x150.jpg 150w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-275-768x769.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-275.jpg 1108w\" sizes=\"auto, (max-width: 1022px) 100vw, 1022px\" \/><\/figure>\n\n\n\n<p><sup>&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; Agentic RL 6 \u5927\u6838\u5fc3\u80fd\u529b\u677f\u5757<\/sup><\/p>\n\n\n\n<p><strong>\u4efb\u52a1\u843d\u5730\u4e0e\u6f14\u5316\u8def\u5f84<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"632\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-276-1024x632.jpg\" alt=\"\" class=\"wp-image-978\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-276-1024x632.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-276-300x185.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-276-768x474.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-276.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><sup>&nbsp; &nbsp; &nbsp; \u4e0d\u540c\u4efb\u52a1\u9886\u57df\u7684 Agent RL \u8fdb\u5316\u6811<\/sup><\/p>\n\n\n\n<p>Agentic RL \u4e5f\u5728\u6a2a\u5411\u62d3\u5c55\u5e94\u7528\u8fb9\u754c\uff0c\u6d8c\u73b0\u51fa\u591a\u79cd\u667a\u80fd\u4f53\u96cf\u5f62\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u641c\u7d22\u4e0e\u7814\u7a76<\/strong>\uff1a\u4f18\u5316\u591a\u8f6e\u68c0\u7d22\u4e0e\u8bc1\u636e\u6574\u5408\u7b56\u7565\uff0c\u5b66\u4f1a\u4f55\u65f6\u7ee7\u7eed\u641c\u7d22\u3001\u4f55\u65f6\u4e0b\u7ed3\u8bba\uff1b<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"674\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-277-1024x674.jpg\" alt=\"\" class=\"wp-image-979\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-277-1024x674.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-277-300x197.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-277-768x505.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-277.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u4ee3\u7801<\/strong>\uff1a\u5c06\u7f16\u8bd1\u9519\u8bef\u4e0e\u5355\u5143\u6d4b\u8bd5\u7ed3\u679c\u7528\u4f5c\u5956\u52b1\uff0c\u63a8\u52a8\u667a\u80fd\u4f53\u80fd\u529b\u4ece\u4e00\u6b21\u6027\u4ee3\u7801\u751f\u6210\u8fdb\u5316\u5230\u81ea\u52a8\u8c03\u8bd5\u4ee5\u53ca\u81ea\u52a8\u5316\u8f6f\u4ef6\u5de5\u7a0b\u6d41\u7a0b\uff1b<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1005\" height=\"1024\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-278-1005x1024.jpg\" alt=\"\" class=\"wp-image-980\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-278-1005x1024.jpg 1005w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-278-294x300.jpg 294w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-278-768x783.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-278.jpg 1108w\" sizes=\"auto, (max-width: 1005px) 100vw, 1005px\" \/><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u6570\u5b66<\/strong>\uff1a\u5728\u975e\u5f62\u5f0f\u5316 (informal) \u63a8\u7406\u4e2d\uff0c\u5229\u7528\u6b63\u786e\u7387\u6216\u4e2d\u95f4\u8fc7\u7a0b\u5956\u52b1\u6765\u5851\u9020\u63a8\u7406\u8f68\u8ff9\uff1b\u5728\u5f62\u5f0f\u5316 (formal) \u63a8\u7406\u4e2d\uff0c\u4ea4\u4e92\u5f0f\u5b9a\u7406\u8bc1\u660e\u5668 (ITPs) \u63d0\u4f9b\u53ef\u9a8c\u8bc1\u7684\u4e8c\u503c\u4fe1\u53f7\uff0c\u4f7f\u667a\u80fd\u4f53\u80fd\u5728\u4e25\u683c\u89c4\u5219\u4e0b\u63a2\u7d22\u8bc1\u660e\u8def\u5f84\uff1b<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"901\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-279-1024x901.jpg\" alt=\"\" class=\"wp-image-981\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-279-1024x901.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-279-300x264.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-279-768x676.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-279.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u56fe\u5f62\u754c\u9762 (GUI)<\/strong>\uff1a\u5728\u7f51\u9875\u548c\u684c\u9762\u73af\u5883\u4e2d\u8ba9\u667a\u80fd\u4f53\u5b66\u4e60\u70b9\u51fb\u3001\u8f93\u5165\u3001\u5bfc\u822a\u7b49\u64cd\u4f5c\uff0c\u4ece\u9759\u6001\u811a\u672c\u6a21\u4eff\u8d70\u5411\u4ea4\u4e92\u5f0f\u64cd\u4f5c\uff0c\u63d0\u5347\u5bf9\u771f\u5b9e\u5e94\u7528\u7684\u9002\u914d\u6027\uff1b<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"687\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-280-1024x687.jpg\" alt=\"\" class=\"wp-image-982\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-280-1024x687.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-280-300x201.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-280-768x515.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-280.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u89c6\u89c9\u4e0e\u5177\u8eab<\/strong>\uff1a\u878d\u5408\u89c6\u89c9\u611f\u77e5\u4e0e\u51b3\u7b56\u89c4\u5212\uff0c\u5b9e\u73b0\u300c\u770b-\u601d-\u505a\u300d\u7684\u8fde\u7eed\u51b3\u7b56\u95ed\u73af\uff0c\u589e\u5f3a\u667a\u80fd\u4f53\u5728\u591a\u6a21\u6001\u95ee\u7b54\u3001\u5bfc\u822a\u4e0e\u673a\u5668\u4eba\u64cd\u4f5c\u7b49\u4efb\u52a1\u4e2d\u7684\u8868\u73b0\uff1b<\/li>\n\n\n\n<li><strong>\u591a\u667a\u80fd\u4f53\u7cfb\u7edf<\/strong>\uff1a\u901a\u8fc7\u5956\u52b1\u8bbe\u8ba1\u4fc3\u4f7f\u591a\u4e2a\u6a21\u578b\u5728\u7ade\u4e89\u6216\u5408\u4f5c\u4e2d\u9010\u6e10\u6d8c\u73b0\u6c9f\u901a\u4e0e\u5206\u5de5\u80fd\u529b\u3002<\/li>\n<\/ul>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"769\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-281-1024x769.jpg\" alt=\"\" class=\"wp-image-983\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-281-1024x769.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-281-300x225.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-281-768x577.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-281.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u5176\u4ed6\u63a2\u7d22<\/strong>\uff1aRL \u4e5f\u88ab\u5e94\u7528\u4e8e\u6570\u636e\u5206\u6790\u3001\u79d1\u5b66\u53d1\u73b0\u7b49\u573a\u666f\uff0c\u663e\u793a\u51fa Agentic RL \u5728\u66f4\u591a\u4efb\u52a1\u4e2d\u7684\u6f5c\u5728\u9002\u5e94\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u603b\u4f53\u6765\u770b\uff0cAgentic RL \u5df2\u5728\u591a\u4e2a\u573a\u666f\u521d\u6b65\u843d\u5730\uff0c\u5e76\u6b63\u4ece\u5355\u4e00\u4efb\u52a1\u9010\u6e10\u8fc8\u5411\u66f4\u590d\u6742\u3001\u66f4\u8d34\u8fd1\u73b0\u5b9e\u7684\u4efb\u52a1\u751f\u6001\u3002<\/p>\n\n\n\n<p><strong>\u73af\u5883\u4e0e\u6846\u67b6<\/strong><\/p>\n\n\n\n<p>Agentic RL \u7684\u53d1\u5c55\u79bb\u4e0d\u5f00\u53ef\u590d\u7528\u7684\u5b9e\u9a8c\u73af\u5883\u4e0e\u5de5\u5177\u94fe\u3002\u73b0\u6709\u5de5\u4f5c\u5df2\u6db5\u76d6\u7f51\u9875\u3001GUI\u3001\u4ee3\u7801\u3001\u6e38\u620f\u7b49\u591a\u79cd\u5f00\u6e90\u5e73\u53f0\uff0c\u5e76\u914d\u5957\u4e86\u76f8\u5e94\u7684\u8bc4\u6d4b\u57fa\u51c6\u4e0e\u6846\u67b6\uff0c\u4e3a\u7814\u7a76\u8005\u63d0\u4f9b\u4e86\u5f00\u5c55\u5b9e\u9a8c\u548c\u5bf9\u6bd4\u7684\u57fa\u7840\u8bbe\u65bd\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"888\" height=\"1024\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-282-888x1024.jpg\" alt=\"\" class=\"wp-image-984\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-282-888x1024.jpg 888w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-282-260x300.jpg 260w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-282-768x886.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-282.jpg 1108w\" sizes=\"auto, (max-width: 888px) 100vw, 888px\" \/><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"758\" src=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-283-1024x758.jpg\" alt=\"\" class=\"wp-image-985\" srcset=\"http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-283-1024x758.jpg 1024w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-283-300x222.jpg 300w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-283-768x568.jpg 768w, http:\/\/www.applicationsofllm.com\/wp-content\/uploads\/2025\/09\/image-283.jpg 1108w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p>\u6b64\u5916\uff0c\u8fd9\u4efd\u7efc\u8ff0\u8fd8\u6574\u5408\u4e86 500+ \u7bc7\u76f8\u5173\u7814\u7a76\uff0c\u5e76\u5728 GitHub \u4e0a\u5f00\u6e90\u4e86 Awesome-AgenticLLM-RL-Papers\uff0c\u5c06\u8bba\u6587\u3001\u73af\u5883\u3001\u57fa\u51c6\u4e0e\u6846\u67b6\u4e00\u7ad9\u5f0f\u6c47\u603b\uff0c\u4e3a\u540e\u7eed\u7814\u7a76\u63d0\u4f9b\u4e86\u5168\u666f\u5f0f\u7684\u53c2\u8003\u5730\u56fe\u3002<\/p>\n\n\n\n<p><strong>\u6311\u6218\u4e0e\u524d\u77bb<\/strong><\/p>\n\n\n\n<p>\u5c3d\u7ba1 Agentic RL \u5df2\u5c55\u73b0\u51fa\u5e7f\u9614\u6f5c\u529b\uff0c\u4f46\u8981\u771f\u6b63\u8d70\u5411\u7a33\u5065\u548c\u5b9e\u7528\uff0c\u8fd8\u5b58\u5728\u82e5\u5e72\u6838\u5fc3\u6311\u6218\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u53ef\u4fe1\u6027\u4e0e\u5b89\u5168\u6027<\/strong>\uff1a\u76f8\u6bd4\u4f20\u7edf LLM\uff0cAgentic RL \u667a\u80fd\u4f53\u96c6\u6210\u4e86\u89c4\u5212\u3001\u5de5\u5177\u8c03\u7528\u548c\u8bb0\u5fc6\u7b49\u80fd\u529b\uff0c\u653b\u51fb\u9762\u663e\u8457\u6269\u5927\uff1b\u540c\u65f6\uff0cRL \u7684\u5956\u52b1\u9a71\u52a8\u673a\u5236\u4e5f\u53ef\u80fd\u5bfc\u81f4 reward hacking\uff0c\u4f7f\u4e0d\u5b89\u5168\u884c\u4e3a\u88ab\u5f3a\u5316\uff0c\u5e26\u6765\u66f4\u6301\u4e45\u7684\u98ce\u9669\u3002<\/li>\n\n\n\n<li><strong>Scale Up \u667a\u80fd\u4f53\u8bad\u7ec3<\/strong>\uff1a\u5927\u89c4\u6a21 Agentic RL \u8bad\u7ec3\u9762\u4e34\u7b97\u529b\u3001\u6570\u636e\u548c\u7b97\u6cd5\u6548\u7387\u7684\u74f6\u9888\u3002\u5f53\u524d RL \u65b9\u6cd5\u6210\u672c\u9ad8\u6602\uff0c\u96be\u4ee5\u5728\u957f\u65f6\u7a0b\u51b3\u7b56\u6216\u590d\u6742\u73af\u5883\u4e2d\u7a33\u5b9a\u6269\u5c55\uff0c\u9700\u8981\u53d1\u5c55\u66f4\u9ad8\u6548\u7684\u4f18\u5316\u8303\u5f0f\u3002<\/li>\n\n\n\n<li><strong>Scale Up \u667a\u80fd\u4f53\u73af\u5883<\/strong>\uff1a\u73b0\u6709\u7684\u4ea4\u4e92\u73af\u5883\u96be\u4ee5\u8986\u76d6\u771f\u5b9e\u4e16\u754c\u7684\u590d\u6742\u6027\u3002\u672a\u6765\u5e94\u63a2\u7d22\u73af\u5883\u4e0e\u667a\u80fd\u4f53\u7684\u300c\u534f\u540c\u8fdb\u5316\u300d\uff0c\u4f8b\u5982\u901a\u8fc7\u81ea\u52a8\u5316\u5956\u52b1\u8bbe\u8ba1\u3001\u8bfe\u7a0b\u751f\u6210\u548c\u73af\u5883\u81ea\u9002\u5e94\u4f18\u5316\uff0c\u8ba9\u73af\u5883\u5728\u8bad\u7ec3\u4e2d\u53d1\u6325\u300c\u4e3b\u52a8\u6559\u5b66\u300d\u7684\u4f5c\u7528\uff0c\u800c\u4e0d\u4ec5\u4ec5\u4f5c\u4e3a\u9759\u6001\u6d4b\u8bd5\u5e73\u53f0\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u8fd9\u4e9b\u6311\u6218\u6784\u6210\u4e86 Agentic RL \u8fdb\u4e00\u6b65\u53d1\u5c55\u7684\u5173\u952e\u95e8\u69db\uff0c\u4e5f\u4e3a\u672a\u6765\u7814\u7a76\u63d0\u4f9b\u4e86\u660e\u786e\u65b9\u5411\u3002<\/p>\n\n\n\n<p><strong>\u7ed3\u8bed<\/strong><\/p>\n\n\n\n<p>\u8fd9\u7bc7\u7efc\u8ff0\u7cfb\u7edf\u5316\u68b3\u7406\u4e86 Agentic RL \u7684\u7406\u8bba\u6846\u67b6\u3001\u80fd\u529b\u7ef4\u5ea6\u3001\u4efb\u52a1\u5e94\u7528\u4e0e\u8d44\u6e90\u751f\u6001\uff0c\u786e\u7acb\u4e86\u5176\u4f5c\u4e3a LLM \u6f14\u8fdb\u7684\u91cd\u8981\u8bad\u7ec3\u8303\u5f0f\u3002<\/p>\n\n\n\n<p>\u7efc\u8ff0\u5f3a\u8c03\uff1a\u5355\u6b65\u5bf9\u9f50\u5df2\u96be\u4ee5\u652f\u6491\u590d\u6742\u4efb\u52a1\uff0cLLM \u8bad\u7ec3\u8303\u5f0f\u7531\u6b64\u8fdb\u5165 Agentic RL \u7684\u4e0b\u534a\u573a\uff0c\u800c\u5f3a\u5316\u5b66\u4e60\u662f\u5c06\u89c4\u5212\u3001\u5de5\u5177\u4f7f\u7528\u3001\u8bb0\u5fc6\u3001\u63a8\u7406\u7b49\u6838\u5fc3\u80fd\u529b\u4ece\u542f\u53d1\u5f0f\u529f\u80fd\u8f6c\u5316\u4e3a\u7a33\u5065\u667a\u80fd\u884c\u4e3a\u7684\u5173\u952e\u673a\u5236\u3002<\/p>\n\n\n\n<p>\u672a\u6765\uff0c\u968f\u7740\u53ef\u4fe1\u6027\u3001\u53ef\u6269\u5c55\u6027\u548c\u590d\u6742\u73af\u5883\u7b49\u6311\u6218\u7684\u9010\u6b65\u7a81\u7834\uff0cLLM \u5c06\u6709\u671b\u771f\u6b63\u4ece \u300c\u4f1a\u8bf4\u300d\u8fc8\u5411\u300c\u4f1a\u505a\u300d\uff0c\u6210\u957f\u4e3a\u66f4\u901a\u7528\u3001\u66f4\u81ea\u4e3b\u7684\u667a\u80fd\u4f53\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u8fc7\u53bb\u51e0\u5e74\uff0c\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u7684\u8bad\u7ec3\u5927\u591a\u4f9d\u8d56\u4e8e\u57fa\u4e8e\u4eba\u7c7b\u6216\u6570\u636e\u504f\u597d\u7684\u5f3a\u5316\u5b66\u4e60\uff08Preference-based  [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":968,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[10],"tags":[],"class_list":["post-967","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-10"],"_links":{"self":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts\/967","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/comments?post=967"}],"version-history":[{"count":1,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts\/967\/revisions"}],"predecessor-version":[{"id":986,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/posts\/967\/revisions\/986"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/media\/968"}],"wp:attachment":[{"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/media?parent=967"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/categories?post=967"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.applicationsofllm.com\/index.php\/wp-json\/wp\/v2\/tags?post=967"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}