{"id":22305,"date":"2024-09-23T11:26:53","date_gmt":"2024-09-23T03:26:53","guid":{"rendered":"https:\/\/aif.amtbbs.org\/?p=22305"},"modified":"2024-09-23T11:26:53","modified_gmt":"2024-09-23T03:26:53","slug":"%e5%bc%ba%e5%8c%96%e5%ad%a6%e4%b9%a0%e8%ae%a9%e5%a4%a7%e6%a8%a1%e5%9e%8b%e8%87%aa%e5%8a%a8%e7%ba%a0%e9%94%99%ef%bc%8c%e6%95%b0%e5%ad%a6%e3%80%81%e7%bc%96%e7%a8%8b%e6%80%a7%e8%83%bd%e6%9a%b4%e6%b6%a8","status":"publish","type":"post","link":"https:\/\/aif.amtbbs.org\/index.php\/2024\/09\/23\/22305\/","title":{"rendered":"\u5f3a\u5316\u5b66\u4e60\u8ba9\u5927\u6a21\u578b\u81ea\u52a8\u7ea0\u9519\uff0c\u6570\u5b66\u3001\u7f16\u7a0b\u6027\u80fd\u66b4\u6da8\uff0cDeepMind\u65b0\u4f5c"},"content":{"rendered":"<div><img data-dominant-color=\"a9b8c6\" data-has-transparency=\"false\" style=\"--dominant-color: #a9b8c6;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-full wp-image-22308\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/09\/9859a5d816212c27d771211d3aeb4233f04810-300x167-1.jpg\" width=\"300\" height=\"167\" alt=\"\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/09\/9859a5d816212c27d771211d3aeb4233f04810-300x167-1.jpg 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/09\/9859a5d816212c27d771211d3aeb4233f04810-300x167-1-150x84.jpg 150w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" \/><\/div>\n<div><\/div>\n<div class=\"article-desc\">\u65e0\u9700\u4f9d\u8d56\u5916\u90e8\u53cd\u9988\u6216\u989d\u5916\u6a21\u578b\uff0c\u7eaf\u7eaf\u7684\u81ea\u6211\u7ea0\u6b63\u3002<\/div>\n<div id=\"postspictures\" class=\"article-content\">\n<div id=\"container\" class=\"container am-engine\" data-v-1d7a5742=\"\" data-element=\"root\">\n<p>\u81ea\u6211\u7ea0\u6b63\uff08Self-correction\uff09\u662f\u5927\u8bed\u8a00\u6a21\u578b (LLM) \u975e\u5e38\u91cd\u8981\u7684\u80fd\u529b\uff0c\u4f46\u4eba\u4eec\u53d1\u73b0\u8fd9\u79cd\u80fd\u529b\u5728\u73b0\u4ee3 LLM \u4e2d\u57fa\u672c\u4e0a\u5f88\u5c11\u5b58\u5728\u3002\u73b0\u6709\u7684\u8bad\u7ec3\u81ea\u6211\u7ea0\u6b63\u7684\u65b9\u6cd5\u8981\u4e48\u9700\u8981\u591a\u4e2a\u6a21\u578b\uff0c\u8981\u4e48\u4f9d\u8d56\u4e8e\u66f4\u5f3a\u5927\u7684\u6a21\u578b\u6216\u5176\u4ed6\u5f62\u5f0f\u7684\u76d1\u7763\u3002<\/p>\n<p>\u6211\u4eec\u5982\u4f55\u624d\u80fd\u8ba9 LLM \u5177\u5907\u81ea\u6211\u7ea0\u6b63\u80fd\u529b\uff1f\u4e4b\u524d\u7684\u7814\u7a76\u8981\u4e48\u4f9d\u8d56\u4e8e\u63d0\u793a\u5de5\u7a0b\uff0c\u8981\u4e48\u4f9d\u8d56\u4e8e\u4e13\u95e8\u7528\u4e8e\u81ea\u6211\u7ea0\u6b63\u7684\u5fae\u8c03\u6a21\u578b\u3002\u4f46\u524d\u8005\u901a\u5e38\u65e0\u6cd5\u6709\u6548\u5730\u8fdb\u884c\u6709\u610f\u4e49\u7684\u5185\u5728\u81ea\u6211\u7ea0\u6b63\uff0c\u800c\u540e\u8005\u57fa\u4e8e\u5fae\u8c03\u7684\u65b9\u6cd5\u9700\u8981\u5728\u63a8\u7406\u65f6\u8fd0\u884c\u591a\u4e2a\u6a21\u578b\uff0c\u4f8b\u5982\u9700\u8981 oracle\u300c\u6559\u5e08\u300d\u6765\u76d1\u7763\u6307\u5bfc\u81ea\u6211\u7ea0\u6b63\u8fc7\u7a0b\u3002<\/p>\n<p>\u5728\u6700\u8fd1\u63d0\u4ea4\u7684\u4e00\u7bc7\u8bba\u6587\u4e2d\uff0c\u6765\u81ea Google DeepMind \u7684\u7814\u7a76\u8005\u5f00\u53d1\u4e86\u4e00\u79cd\u65e0\u9700\u4e0a\u8ff0\u4efb\u4f55\u8981\u6c42\u5373\u53ef\u6709\u6548\u8fdb\u884c\u81ea\u6211\u7ea0\u6b63\u7684\u65b9\u6cd5\uff0c\u5373\u901a\u8fc7\u5f3a\u5316\u5b66\u4e60\u8fdb\u884c\u81ea\u6211\u7ea0\u6b63\uff08SCoRe\uff0cSelf-Correction via Reinforcement Learning)\uff0c\u53ea\u9700\u8bad\u7ec3\u4e00\u4e2a\u6a21\u578b\uff0c\u8be5\u6a21\u578b\u65e2\u53ef\u4ee5\u5bf9\u63a8\u7406\u95ee\u9898\u505a\u51fa\u54cd\u5e94\uff0c\u4e5f\u53ef\u4ee5\u7ea0\u6b63\u9519\u8bef\uff0c\u5c3d\u7ba1\u6ca1\u6709\u6536\u5230\u4efb\u4f55 oracle \u53cd\u9988\u3002\u66f4\u91cd\u8981\u7684\u662f\uff0cSCoRe \u5b8c\u5168\u901a\u8fc7\u5728\u81ea\u751f\u6210\u6570\u636e\u4e0a\u8fdb\u884c\u8bad\u7ec3\u6765\u6559\u6a21\u578b\u5177\u5907\u8fd9\u79cd\u80fd\u529b\uff0c\u800c\u65e0\u9700\u4efb\u4f55 oracle\u3002<\/p>\n<p>&nbsp;<\/p>\n<ul data-id=\"ua73dd4b-BL1QcDO2\">\n<li data-id=\"ld70c578-1TA1BlN0\">\u8bba\u6587\u6807\u9898\uff1aTraining Language Models to Self-Correct via\u00a0Reinforcement Learning<\/li>\n<li data-id=\"ld70c578-0jkmDI0D\">\u8bba\u6587\u5730\u5740\uff1ahttps:\/\/arxiv.org\/pdf\/2409.12917<\/li>\n<\/ul>\n<p>\u672c\u6587\u4e3b\u8981\u8d21\u732e\u5728\u4e8e\u63d0\u51fa\u4e86\u4e00\u79cd\u591a\u8f6e\u5f3a\u5316\u5b66\u4e60\u65b9\u6cd5 \u2014\u2014SCoRe\uff0c\u7528\u4e8e\u6559 LLM \u5982\u4f55\u7ea0\u6b63\u81ea\u5df1\u7684\u9519\u8bef\u3002\u76f8\u5bf9\u4e8e\u57fa\u7840 Gemini \u6a21\u578b\uff0cSCoRe \u5728 MATH \u63a8\u7406\u95ee\u9898\u7684\u81ea\u6211\u7ea0\u6b63\u65b9\u9762\u83b7\u5f97\u4e86 15.6% \u7684\u589e\u76ca\uff0c\u5728 HumanEval \u7f16\u7801\u95ee\u9898\u4e0a\u83b7\u5f97\u4e86 9.1% \u7684\u589e\u76ca\u3002<\/p>\n<h4>SCoRe \u539f\u7406\u4ecb\u7ecd<\/h4>\n<p>\u4e3a\u4e86\u6559 LLM \u8fdb\u884c\u81ea\u6211\u7ea0\u6b63\uff0cSCoRe \u5c06\u6807\u51c6\u5355\u8f6e\u5f3a\u5316\u5b66\u4e60\uff08\u516c\u5f0f 2\uff09\u6269\u5c55\u5230 Zhou \u7b49\u4eba\u63d0\u51fa\u7684\u5206\u5c42\u6846\u67b6\u4e0b\u7684\u591a\u8f6e\u8bbe\u7f6e\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s7.51cto.com\/oss\/202409\/22\/f13dca58361a44a18b0623114901868e508df1.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u4e0d\u8fc7\u8fd9\u6837\u505a\u9762\u4e34\u8bf8\u591a\u6311\u6218\u3002\u9996\u5148\uff0c\u4f18\u5316\u516c\u5f0f 1 \u89e3\u51b3\u4e86\u5206\u5e03\u504f\u79fb\u95ee\u9898\uff0c\u4f46\u5c1a\u4e0d\u6e05\u695a\u5b83\u662f\u5426\u4e5f\u80fd\u6ee1\u8db3\u8981\u6c42 [D2]\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s8.51cto.com\/oss\/202409\/22\/a6932770645293fdc9c595d2738b25bb6d225e.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u8fd9\u91cc\u7684 [D2] \u5982\u4e0b\u56fe\u6240\u793a\uff0c\u56fe\u4e2d\u5c55\u793a\u4e86 SFT \u65b9\u6cd5\u5931\u8d25\u7684\u4e24\u4e2a\u539f\u56e0\u3002\u800c\u6709\u6548\u7684\u89e3\u51b3\u65b9\u6848\u5fc5\u987b\u6ee1\u8db3\u4e24\u4e2a\u8981\u6c42\uff1a[D1] \u6a21\u578b\u5e94\u8be5\u76f4\u63a5\u5728\u81ea\u751f\u6210\u8f68\u8ff9\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u4ee5\u7f13\u89e3 SFT \u7684\u5206\u5e03\u4e0d\u5339\u914d\uff08\u56fe 4\uff09\uff0c[D2] \u6240\u91c7\u7528\u7684\u81ea\u751f\u6210\u8f68\u8ff9\u5e94\u9632\u6b62\u5728\u5b66\u4e60\u8fc7\u7a0b\u4e2d\u56e0\u8fdb\u884c\u5fae\u5c0f\u7f16\u8f91\u800c\u5d29\u6e83\u3002<\/p>\n<p><strong>\u4f5c\u8005\u5f00\u53d1\u4e86\u4e00\u79cd\u5728\u7ebf RL \u65b9\u6cd5\uff0c\u901a\u8fc7\u4ed4\u7ec6\u7684\u521d\u59cb\u5316\u548c\u5956\u52b1\u5851\u9020\u6765\u89e3\u51b3\u8fd9\u4e9b\u6311\u6218\u3002<\/strong><\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s6.51cto.com\/oss\/202409\/22\/937479358c7b7b1fecf5136442580f047bff21.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s6.51cto.com\/oss\/202409\/22\/e20e883287650498a96222c1e98422e287c1d2.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u5176\u6b21\u7528\u4e8e\u5fae\u8c03\u7684\u57fa\u7840\u6a21\u578b\u521d\u59cb\u5316\u5728\u7f16\u8f91\u8ddd\u79bb\u4e0a\u5448\u73b0\u51fa\u9ad8\u5ea6\u503e\u659c\u7684\u5206\u5e03\uff08\u56fe 3a\uff09\uff0c\u8fd9\u4f7f\u5f97\u5b83\u4eec\u5bb9\u6613\u53d7\u5230\u6a21\u5f0f\u5d29\u6e83\u7684\u5f71\u54cd\uff0c\u8fd9\u662f\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u4e2d\u4e00\u4e2a\u5e38\u89c1\u7684\u95ee\u9898\u3002\u5373\u4f7f\u57fa\u7840\u6a21\u578b\u53ef\u4ee5\u5728\u81ea\u6211\u6821\u6b63\u8fc7\u7a0b\u4e2d\u4ea7\u751f\u7f16\u8f91\u8ddd\u79bb\u6bd4\u503e\u659c\u5ea6\u8f83\u5c0f\u7684\u5206\u5e03\uff0c\u4f46\u4ecd\u7136\u9700\u8981\u5f3a\u5316\u5b66\u4e60\u8bad\u7ec3\u8fc7\u7a0b\u4ece\u8bad\u7ec3\u6570\u636e\u4e2d\u5b66\u4e60\u4e00\u79cd\u53ef\u4ee5\u63a8\u5e7f\u5230\u6d4b\u8bd5\u63d0\u793a\u7684\u81ea\u6211\u6821\u6b63\u7b56\u7565\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s2.51cto.com\/oss\/202409\/22\/658f67521bfcd49dd09009a221aed8928e1057.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>SCoRe \u65e8\u5728\u89e3\u51b3\u4e0a\u8ff0\u5173\u952e\u6311\u6218\uff0c\u5176\u5206\u4e3a\u4e24\u4e2a\u9636\u6bb5\uff0c\u8fd9\u4e24\u4e2a\u9636\u6bb5\u90fd\u662f\u901a\u8fc7\u9002\u5f53\u521d\u59cb\u5316\u6a21\u578b\u548c\u63a7\u5236\u540e\u7eed RL \u4f7f\u6a21\u578b\u504f\u5411\u5b66\u4e60\u81ea\u6211\u7ea0\u6b63\u3002<\/p>\n<p>\u5177\u4f53\u800c\u8a00\uff0c\u8fd9\u4e24\u4e2a\u9636\u6bb5\u5305\u62ec\uff1a<\/p>\n<p><strong>\u9636\u6bb5 I\uff1a\u8bad\u7ec3\u6a21\u578b\u521d\u59cb\u5316\u4ee5\u9632\u6b62\u5d29\u6e83<\/strong><\/p>\n<p>SCoRe \u7b2c\u4e00\u9636\u6bb5\u7684\u76ee\u6807\u662f\u901a\u8fc7\u63d0\u9ad8\u57fa\u7840\u6a21\u578b\u5bf9\u7b2c\u4e8c\u6b21\u5c1d\u8bd5\u54cd\u5e94\u7684\u8986\u76d6\u7387\u6765\u83b7\u5f97\u826f\u597d\u7684\u6a21\u578b\u521d\u59cb\u5316\uff0c\u4ee5\u4fbf\u540e\u7eed\u81ea\u6211\u7ea0\u6b63\u8bad\u7ec3\u4e0d\u4f1a\u51fa\u73b0 STaR\/SFT \u4e2d\u89c2\u5bdf\u5230\u7684\u5d29\u6e83\u73b0\u8c61\u3002<\/p>\n<p>\u4e3a\u4e86\u8fbe\u5230\u6b64\u76ee\u7684\uff0c\u8be5\u7814\u7a76\u4e0d\u91c7\u7528 SFT \u6765\u521d\u59cb\u5316 RL \u8bad\u7ec3\uff0c\u800c\u662f\u5f00\u53d1\u4e86\u7b2c\u4e00\u9636\u6bb5\u6765\u4ea7\u751f\u4e0d\u6613\u5d29\u6e83\u7684\u5355\u72ec\u521d\u59cb\u5316\u3002<\/p>\n<p>\u4f5c\u8005\u5fae\u8c03\u57fa\u7840\u6a21\u578b\uff0c\u4ee5\u4fbf\u5728\u7b2c\u4e8c\u6b21\u5c1d\u8bd5\u65f6\u4ea7\u751f\u9ad8\u5956\u52b1\u4fee\u6b63\uff0c\u540c\u65f6\u901a\u8fc7\u4f7f\u7528 KL \u6563\u5ea6\u5c06\u7b2c\u4e00\u6b21\u5c1d\u8bd5\u7684\u54cd\u5e94\u5206\u5e03\u9650\u5236\u4e3a\u5c3d\u53ef\u80fd\u63a5\u8fd1\u57fa\u7840\u6a21\u578b\u7684\u54cd\u5e94\u5206\u5e03\uff0c\u4ece\u800c\u5f3a\u5236\u6a21\u578b\u4e0d\u6539\u53d8\u5176\u7b2c\u4e00\u6b21\u5c1d\u8bd5\u7684\u54cd\u5e94\u3002\u867d\u7136\u8fd9\u770b\u8d77\u6765\u4e0d\u662f\u6700\u4f18\u7684 \u2014\u2014 \u4f46\u7b2c\u4e00\u6b21\u5c1d\u8bd5\u7684\u54cd\u5e94\u9519\u8bef\u8f83\u5c11\uff0c\u53ef\u4ee5\u7ea0\u6b63\u4e3a\u66f4\u597d\u7684\u7b2c\u4e8c\u6b21\u5c1d\u8bd5\u54cd\u5e94\u3002\u4f18\u5316\u7684\u76ee\u6807\u53ef\u4ee5\u8868\u793a\u4e3a\uff1a<\/p>\n<p>&nbsp;<\/p>\n<p>\u5176\u4e2d \ud835\udefd_2 \u662f\u4e00\u4e2a\u8d85\u53c2\u6570\uff0c\u65e8\u5728\u4ec5\u5728\u7b2c\u4e00\u6b21\u5c1d\u8bd5\u65f6\u5f3a\u5236\u6267\u884c\u4e25\u683c\u7684 KL \u60e9\u7f5a\uff0c\u4ee5\u907f\u514d\u7b2c\u4e00\u8f6e\u54cd\u5e94\u53d1\u751f\u504f\u79fb\uff08\u7528\u84dd\u8272\u9879\u8868\u793a\uff09\u3002\u8bf7\u6ce8\u610f\uff0c\u4f5c\u8005\u4ecd\u7136\u4f7f\u7528\u516c\u5f0f 2 \u4e2d\u7684\u9ed8\u8ba4 KL \u6563\u5ea6\u60e9\u7f5a\uff0c\u4f46\u8be5\u60e9\u7f5a\u7684\u6743\u91cd\u8981\u5c0f\u5f97\u591a\uff0c\u5e76\u4e14\u4e3a\u4e86\u7b80\u6d01\u8d77\u89c1\uff0c\u516c\u5f0f 3 \u4e2d\u7701\u7565\u4e86\u5b83\u3002\u4e8b\u5b9e\u4e0a\uff0c\u4e0e\u7b80\u5355\u7684\u591a\u8f6e RL \u4e0d\u540c\uff0c\u9636\u6bb5 I \u5728\u5206\u79bb\u4e24\u4e2a\u54cd\u5e94\u65b9\u9762\u66f4\u6709\u6548\uff08\u56fe 5b\uff09\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s5.51cto.com\/oss\/202409\/22\/7118f523964f2ef7c7306563c28c2ded1fc6f0.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p><strong>\u9636\u6bb5 II\uff1a\u5e26\u6709\u5956\u52b1\u7684\u591a\u8f6e\u5f3a\u5316\u5b66\u4e60<\/strong><\/p>\n<p>\u501f\u52a9\u7b2c\u4e00\u9636\u6bb5\u7684\u6a21\u578b\u521d\u59cb\u5316\uff0c\u8be5\u6a21\u578b\u5728\u8026\u5408\u4e24\u4e2a\u54cd\u5e94\u65f6\u8868\u73b0\u51fa\u66f4\u5c0f\u7684\u504f\u5dee\uff0cSCoRe \u7684\u7b2c\u4e8c\u9636\u6bb5\u73b0\u5728\u53ef\u4ee5\u8bad\u7ec3\u4e24\u6b21\u5c1d\u8bd5\u7684\u54cd\u5e94\uff0c\u5e76\u6839\u636e\u516c\u5f0f 1 \u4f18\u5316\u5956\u52b1\u3002\u5f53\u7136\uff0c\u4f5c\u8005\u8fd8\u5e0c\u671b\u786e\u4fdd\u5728\u6b64\u8fc7\u7a0b\u4e2d\u4e0d\u4f1a\u964d\u4f4e\u7b2c\u4e00\u6b21\u5c1d\u8bd5\u7684\u54cd\u5e94\u3002\u56e0\u6b64\uff0c\u5bf9\u4e8e\u4e24\u8f6e\u81ea\u6211\u7ea0\u6b63\u95ee\u9898\uff0c\u4f5c\u8005\u9488\u5bf9\u4ee5\u4e0b\u76ee\u6807\u8bad\u7ec3\u7b56\u7565 \ud835\udf0b_\ud835\udf03(\u22c5\u2223\u22c5)\uff1a<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s7.51cto.com\/oss\/202409\/22\/e960c2681ed8c8a2096355a86ce581426d3950.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u56fe 6 \u4e3a\u9636\u6bb5 I\u3001\u9636\u6bb5 II \u6d41\u7a0b\u8bf4\u660e\u3002\u53ef\u4ee5\u770b\u51fa SCoRe \u4ee5\u4ea4\u9519\u65b9\u5f0f\u5e94\u7528\u9636\u6bb5 I \u548c II \u8fdb\u884c\u591a\u6b21\u8fed\u4ee3\u3002<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s4.51cto.com\/oss\/202409\/22\/09042f0108bd99b54991555539e1426be63f0d.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s8.51cto.com\/oss\/202409\/22\/99c4386048573672c84310fac0818b232f0ec5.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<h4>\u5b9e\u9a8c\u8bc4\u4f30<\/h4>\n<p>\u8be5\u7814\u7a76\u8fdb\u884c\u4e86\u4e00\u7cfb\u5217\u5b9e\u9a8c\uff0c\u6765\u9a8c\u8bc1 SCoRe \u5728\u6559 LLM \u5177\u5907\u81ea\u6211\u7ea0\u6b63\u80fd\u529b\u65b9\u9762\u7684\u6709\u6548\u6027\uff0c\u5e76\u901a\u8fc7\u6d88\u878d\u5b9e\u9a8c\u63a2\u7d22\u4e86 SCoRe \u7684\u6bcf\u4e2a\u7ec4\u4ef6\u7684\u5f71\u54cd\u3002<\/p>\n<p>\u8be5\u7814\u7a76\u4e3b\u8981\u5173\u6ce8\u6570\u5b66\u548c\u7f16\u7801\u4efb\u52a1\uff0c\u4f7f\u7528\u4ee5\u4e0b\u57fa\u51c6\u6765\u8bc4\u4f30\u65b9\u6cd5\u7684\u6709\u6548\u6027\uff1a<\/p>\n<ul data-id=\"ua73dd4b-KZSlZLbj\">\n<li data-id=\"ld70c578-HVL9nC7l\">MATH\uff1b<\/li>\n<li data-id=\"ld70c578-dQJCA7V1\">MBPP \u548c HumanEval\u3002<\/li>\n<\/ul>\n<p>\u51e0\u79cd\u65b9\u6cd5\u5728 MATH \u57fa\u51c6\u4e0a\u7684\u5b9e\u9a8c\u8bc4\u4f30\u7ed3\u679c\u5982\u4e0b\u8868 3 \u6240\u793a\uff1a<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s5.51cto.com\/oss\/202409\/22\/c3d86ce31805ec32cb9031e48b3642b67a7242.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p>\u5728\u4ee3\u7801\u751f\u6210\u65b9\u9762\uff0c\u5b9e\u9a8c\u7ed3\u679c\u5982\u4e0b\u8868 4 \u6240\u793a\uff1a<\/p>\n<p><img decoding=\"async\" title=\"\u56fe\u7247\" src=\"https:\/\/s6.51cto.com\/oss\/202409\/22\/53efb5c042b6385471b694735d0a353921b404.webp\" alt=\"\u56fe\u7247\" data-type=\"inline\" \/><\/p>\n<p><strong>\u6d88\u878d\u7814\u7a76<\/strong><\/p>\n<p>\u4e3a\u4e86\u63a2\u7a76\u4ee5\u4e0b\u51e0\u4e2a\u95ee\u9898\uff0c\u8be5\u7814\u7a76\u8fdb\u884c\u4e86\u6d88\u878d\u5b9e\u9a8c\uff1a<\/p>\n<ul data-id=\"ua73dd4b-KPiFI5Z8\">\n<li data-id=\"ld70c578-hfKGSo0V\">\u591a\u8f6e\u8bad\u7ec3\u7684\u91cd\u8981\u6027<\/li>\n<li data-id=\"ld70c578-URQqNWKq\">\u591a\u9636\u6bb5\u8bad\u7ec3\u7684\u91cd\u8981\u6027<\/li>\n<li data-id=\"ld70c578-MhpCWjLM\">\u5956\u52b1\u51fd\u6570\u8bbe\u8ba1\uff08reward shaping\uff09\u7684\u5f71\u54cd<\/li>\n<li data-id=\"ld70c578-Bgt40AO0\">on-policy \u5f3a\u5316\u5b66\u4e60\u7684\u91cd\u8981\u6027<\/li>\n<\/ul>\n<p>\u6d88\u878d\u5b9e\u9a8c\u7ed3\u679c\u5982\u4e0b\u8868 5 \u6240\u793a\uff1a<\/p>\n<p><img data-dominant-color=\"e3e3e3\" data-has-transparency=\"false\" style=\"--dominant-color: #e3e3e3;\" loading=\"lazy\" decoding=\"async\" class=\"not-transparent alignnone size-medium wp-image-22306\" src=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/09\/f80326581fccdb70386995ff23ae199cecbdc8-300x106.webp\" alt=\"\" width=\"300\" height=\"106\" srcset=\"https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/09\/f80326581fccdb70386995ff23ae199cecbdc8-300x106.webp 300w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/09\/f80326581fccdb70386995ff23ae199cecbdc8-1024x361.webp 1024w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/09\/f80326581fccdb70386995ff23ae199cecbdc8-150x53.webp 150w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/09\/f80326581fccdb70386995ff23ae199cecbdc8-768x271.webp 768w, https:\/\/aiforumimage.oss-cn-shanghai.aliyuncs.com\/wp-content\/uploads\/2024\/09\/f80326581fccdb70386995ff23ae199cecbdc8.webp 1080w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" \/><\/p>\n<p>\u611f\u5174\u8da3\u7684\u8bfb\u8005\u53ef\u4ee5\u9605\u8bfb\u8bba\u6587\u539f\u6587\uff0c\u4e86\u89e3\u66f4\u591a\u7814\u7a76\u5185\u5bb9\u3002<\/p>\n<\/div>\n<\/div>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_22305\" class=\"pvc_stats total_only  \" data-element-id=\"22305\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u65e0\u9700\u4f9d\u8d56\u5916\u90e8\u53cd\u9988\u6216\u989d\u5916\u6a21\u578b\uff0c\u7eaf\u7eaf\u7684\u81ea\u6211\u7ea0\u6b63\u3002 \u81ea\u6211\u7ea0\u6b63\uff08Self-correction\uff09\u662f\u5927\u8bed\u8a00\u6a21\u578b (LLM [&hellip;]<\/p>\n<div class=\"pvc_clear\"><\/div>\n<p id=\"pvc_stats_22305\" class=\"pvc_stats total_only  \" data-element-id=\"22305\" style=\"\"><i class=\"pvc-stats-icon medium\" aria-hidden=\"true\"><svg xmlns=\"http:\/\/www.w3.org\/2000\/svg\" version=\"1.0\" viewBox=\"0 0 502 315\" preserveAspectRatio=\"xMidYMid meet\"><g transform=\"translate(0,332) scale(0.1,-0.1)\" fill=\"\" stroke=\"none\"><path d=\"M2394 3279 l-29 -30 -3 -207 c-2 -182 0 -211 15 -242 39 -76 157 -76 196 0 15 31 17 60 15 243 l-3 209 -33 29 c-26 23 -41 29 -80 29 -41 0 -53 -5 -78 -31z\"\/><path d=\"M3085 3251 c-45 -19 -58 -50 -96 -229 -47 -217 -49 -260 -13 -295 52 -53 146 -42 177 20 16 31 87 366 87 410 0 70 -86 122 -155 94z\"\/><path d=\"M1751 3234 c-13 -9 -29 -31 -37 -50 -12 -29 -10 -49 21 -204 19 -94 39 -189 45 -210 14 -50 54 -80 110 -80 34 0 48 6 76 34 21 21 34 44 34 59 0 14 -18 113 -40 219 -37 178 -43 195 -70 221 -36 32 -101 37 -139 11z\"\/><path d=\"M1163 3073 c-36 -7 -73 -59 -73 -102 0 -56 133 -378 171 -413 34 -32 83 -37 129 -13 70 36 67 87 -16 290 -86 209 -89 214 -129 231 -35 14 -42 15 -82 7z\"\/><path d=\"M3689 3066 c-15 -9 -33 -30 -42 -48 -48 -103 -147 -355 -147 -375 0 -98 131 -148 192 -74 13 15 57 108 97 206 80 196 84 226 37 273 -30 30 -99 39 -137 18z\"\/><path d=\"M583 2784 c-38 -19 -67 -74 -58 -113 9 -42 211 -354 242 -373 16 -10 45 -18 66 -18 51 0 107 52 107 100 0 39 -1 41 -124 234 -80 126 -108 162 -133 173 -41 17 -61 16 -100 -3z\"\/><path d=\"M4250 2784 c-14 -9 -74 -91 -133 -183 -95 -150 -107 -173 -107 -213 0 -55 33 -94 87 -104 67 -13 90 8 211 198 130 202 137 225 78 284 -27 27 -42 34 -72 34 -22 0 -50 -8 -64 -16z\"\/><path d=\"M2275 2693 c-553 -48 -1095 -270 -1585 -649 -135 -104 -459 -423 -483 -476 -23 -49 -22 -139 2 -186 73 -142 361 -457 571 -626 285 -228 642 -407 990 -497 242 -63 336 -73 660 -74 310 0 370 5 595 52 535 111 1045 392 1455 803 122 121 250 273 275 326 19 41 19 137 0 174 -41 79 -309 363 -465 492 -447 370 -946 591 -1479 653 -113 14 -422 18 -536 8z m395 -428 c171 -34 330 -124 456 -258 112 -119 167 -219 211 -378 27 -96 24 -300 -5 -401 -72 -255 -236 -447 -474 -557 -132 -62 -201 -76 -368 -76 -167 0 -236 14 -368 76 -213 98 -373 271 -451 485 -162 444 86 934 547 1084 153 49 292 57 452 25z m909 -232 c222 -123 408 -262 593 -441 76 -74 138 -139 138 -144 0 -16 -233 -242 -330 -319 -155 -123 -309 -223 -461 -299 l-81 -41 32 46 c18 26 49 83 70 128 143 306 141 649 -6 957 -25 52 -61 116 -79 142 l-34 47 45 -20 c26 -10 76 -36 113 -56z m-2057 25 c-40 -58 -105 -190 -130 -263 -110 -324 -59 -707 132 -981 25 -35 42 -64 37 -64 -19 0 -241 119 -326 174 -188 122 -406 314 -532 468 l-58 71 108 103 c185 178 428 349 672 473 66 33 121 60 123 61 2 0 -10 -19 -26 -42z\"\/><path d=\"M2375 1950 c-198 -44 -350 -190 -395 -379 -18 -76 -8 -221 19 -290 114 -284 457 -406 731 -260 98 52 188 154 231 260 27 69 37 214 19 290 -38 163 -166 304 -326 360 -67 23 -215 33 -279 19z\"\/><\/g><\/svg><\/i> <img loading=\"lazy\" decoding=\"async\" width=\"16\" height=\"16\" alt=\"Loading\" src=\"https:\/\/aif.amtbbs.org\/wp-content\/plugins\/page-views-count\/ajax-loader-2x.gif\" border=0 \/><\/p>\n<div class=\"pvc_clear\"><\/div>\n","protected":false},"author":56,"featured_media":22308,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3,23,20,80],"tags":[840],"class_list":["post-22305","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-23","category-20","category-80","tag-ai-"],"_links":{"self":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/22305","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/users\/56"}],"replies":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/comments?post=22305"}],"version-history":[{"count":1,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/22305\/revisions"}],"predecessor-version":[{"id":22309,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/posts\/22305\/revisions\/22309"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media\/22308"}],"wp:attachment":[{"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/media?parent=22305"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/categories?post=22305"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aif.amtbbs.org\/index.php\/wp-json\/wp\/v2\/tags?post=22305"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}