{"id":38237,"date":"2025-02-12T11:00:55","date_gmt":"2025-02-12T03:00:55","guid":{"rendered":"https:\/\/17aitech.com\/?p=38237"},"modified":"2025-02-12T11:00:55","modified_gmt":"2025-02-12T03:00:55","slug":"%e7%bd%91%e4%bc%a0deepseek-r1%e6%9b%b4%e5%ae%b9%e6%98%93%e8%a2%ab%e8%b6%8a%e7%8b%b1%ef%bc%9f%e8%bf%99%e6%9c%89%e4%b8%aa%e5%85%a5%e9%80%89%e9%a1%b6%e4%bc%9a%e7%9a%84%e9%98%b2%e5%be%a1%e6%a1%86%e6%9e%b6","status":"publish","type":"post","link":"https:\/\/17aitech.com\/?p=38237","title":{"rendered":"\u7f51\u4f20DeepSeek R1\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\uff1f\u8fd9\u6709\u4e2a\u5165\u9009\u9876\u4f1a\u7684\u9632\u5fa1\u6846\u67b6SelfDefend"},"content":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-11-4\" target=\"_blank\">\u7f51\u4f20DeepSeek R1\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\uff1f\u8fd9\u6709\u4e2a\u5165\u9009\u9876\u4f1a\u7684\u9632\u5fa1\u6846\u67b6SelfDefend<\/a><\/p>\n<p><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f9a9ecb29e93345299a082069b54dbc5.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f9a9ecb29e93345299a082069b54dbc5.png\"><\/a><\/p>\n<blockquote data-author-name=\"\" data-content-utf8-length=\"166\" data-source-title=\"\" data-type=\"2\" data-url=\"\">\n<section>\n<p>AIxiv\u4e13\u680f\u662f<mark data-type=\"institutions\" data-id=\"9e96434f-5827-41e3-9b3c-36d39bc0d446\">\u673a\u5668\u4e4b\u5fc3<\/mark>\u53d1\u5e03\u5b66\u672f\u3001\u6280\u672f\u5185\u5bb9\u7684\u680f\u76ee\u3002\u8fc7\u53bb\u6570\u5e74\uff0c<mark data-type=\"institutions\" data-id=\"9e96434f-5827-41e3-9b3c-36d39bc0d446\">\u673a\u5668\u4e4b\u5fc3<\/mark>AIxiv\u4e13\u680f\u63a5\u6536\u62a5\u9053\u4e862000\u591a\u7bc7\u5185\u5bb9\uff0c\u8986\u76d6\u5168\u7403\u5404\u5927\u9ad8\u6821\u4e0e\u4f01\u4e1a\u7684\u9876\u7ea7\u5b9e\u9a8c\u5ba4\uff0c\u6709\u6548\u4fc3\u8fdb\u4e86\u5b66\u672f\u4ea4\u6d41\u4e0e\u4f20\u64ad\u3002\u5982\u679c\u60a8\u6709\u4f18\u79c0\u7684\u5de5\u4f5c\u60f3\u8981\u5206\u4eab\uff0c\u6b22\u8fce\u6295\u7a3f\u6216\u8005\u8054\u7cfb\u62a5\u9053\u3002\u6295\u7a3f\u90ae\u7bb1\uff1aliyazhou@jiqizhixin.com\uff1bzhaoyunfeng@jiqizhixin.com<\/p>\n<\/section>\n<\/blockquote>\n<p><strong>\u672c\u6587\u4e00\u4f5c\u738b\u52cb\u5e7f\u662f\u9999\u6e2f\u79d1\u6280\u5927\u5b66\u7684\u5728\u8bfb\u535a\u58eb\u751f\uff0c\u672c\u79d1\u548c\u7855\u58eb\u5206\u522b\u6bd5\u4e1a\u4e8e\u4e2d\u56fd\u5730\u8d28\u5927\u5b66\u548c\u54c8\u5c14\u6ee8\u5de5\u4e1a\u5927\u5b66\uff0c\u4e3b\u8981\u7814\u7a76\u65b9\u5411\u662f\u5927\u6a21\u578b\u5b89\u5168\u3002<\/strong><strong>\u901a\u8baf\u4f5c\u8005\u5434\u9053\u8fdc\uff0c\u9999\u6e2f\u79d1\u6280\u5927\u5b66\u7814\u7a76\u52a9\u7406\u6559\u6388\uff0c\u7814\u7a76\u65b9\u5411\u5305\u62ec\u5927\u6a21\u578b\u5b89\u5168\u3001<mark data-type=\"tech_methods\" data-id=\"011d6ee9-3cda-4343-935a-e9f3b8eed43a\">\u533a\u5757\u94fe<\/mark>\u548c\u667a\u80fd\u5408\u7ea6\u5b89\u5168\u3001\u79fb\u52a8\u7cfb\u7edf\u548c\u8f6f\u4ef6\u5b89\u5168\u3002<\/strong><strong>\u901a\u8baf\u4f5c\u8005\u738b\u5e05\uff0c\u9999\u6e2f\u79d1\u6280\u5927\u5b66\u957f\u8058\u526f\u6559\u6388\u3002\u7814\u7a76\u65b9\u5411\u5305\u62ec AI \u5b89\u5168\u3001\u8f6f\u4ef6\u5b89\u5168\u3001\u6570\u636e\u9690\u79c1\u3001<mark data-type=\"tech_methods\" data-id=\"60715cba-d78d-45e4-9f5c-bfdbe702ca0e\">\u9006\u5411\u5de5\u7a0b<\/mark>\u7b49\u3002<\/strong><\/p>\n<p>\u6700\u8fd1\u4e00\u6bb5\u65f6\u95f4\uff0cDeepSeek \u53ef\u8c13\u662f\u98ce\u5934\u65e0\u4e24\u3002<\/p>\n<p>\u5728\u5927\u5bb6\u7eb7\u7eb7\u8d5e\u626c\u5176\u8d85\u5f3a\u6027\u80fd\u7684\u540c\u65f6\uff0c\u4e5f\u6709\u5a92\u4f53\u66dd\u51fa DeepSeek \u7684 R1 \u6bd4\u5176\u4ed6 AI \u6a21\u578b\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\u3002<\/p>\n<p>\u6bd4\u5982\uff0c\u6b64\u524d\u5bbe\u5915\u6cd5\u5c3c\u4e9a\u5927\u5b66\u7684\u7814\u7a76\u8005\u4f7f\u7528\u6765\u81eaHarmBench\u6570\u636e\u96c6\u768450\u4e2a\u6709\u5bb3\u63d0\u793a\u5bf9DeepSeek R1\u8fdb\u884c\u6d4b\u8bd5\uff0c\u8fd9\u4e9b\u63d0\u793a\u6db5\u76d6\u7f51\u7edc\u72af\u7f6a\u3001\u865a\u5047\u4fe1\u606f\u548c\u975e\u6cd5\u6d3b\u52a8\u7b49\u9886\u57df\u3002\u7ed3\u679c\u663e\u793a\uff0cDeepSeek\u672a\u80fd\u62e6\u622a\u4efb\u4f55\u4e00\u4e2a\u6709\u5bb3\u8bf7\u6c42\uff0c\u653b\u51fb\u6210\u529f\u7387\u8fbe\u5230\u60ca\u4eba\u7684100%\u3002<\/p>\n<p>\u8fd9\u65f6\u5982\u679c\u6709\u4e00\u4e2a AI \u7cfb\u7edf\u80fd\u50cf\u4eba\u7c7b\u4e00\u6837\u5177\u5907\u81ea\u6211\u4fdd\u62a4\u610f\u8bc6\uff0c\u5728\u9762\u5bf9 &#8216; \u6b3a\u9a97 &#8216; \u65f6\u80fd\u591f\u5f53\u673a\u7acb\u65ad\u5730\u8bc6\u7834\u9634\u8c0b \u2014\u2014 \u8fd9\u4e0d\u518d\u662f\u79d1\u5e7b\u7247\u4e2d\u7684\u573a\u666f\u3002<\/p>\n<p>\u8fd1\u65e5\uff0c\u6765\u81ea\u9999\u6e2f\u79d1\u6280\u5927\u5b66\u3001\u5357\u6d0b\u7406\u5de5\u5927\u5b66\u7b49\u673a\u6784\u7684\u7814\u7a76\u56e2\u961f\u6700\u65b0\u6210\u679c\u8ba9\u8fd9\u4e00\u8bbe\u60f3\u6210\u4e3a\u73b0\u5b9e\u3002\u4ed6\u4eec\u63d0\u51fa\u7684 SelfDefend \u6846\u67b6\uff0c\u8ba9\u5927<mark data-type=\"tech_tasks\" data-id=\"fee178e8-ee20-42fc-8f8f-1d41c1f34e8f\">\u8bed\u8a00\u6a21\u578b<\/mark>\u9996\u6b21\u62e5\u6709\u4e86\u771f\u6b63\u610f\u4e49\u4e0a\u7684 &#8216; \u81ea\u536b\u80fd\u529b &#8216;\uff0c\u80fd\u591f\u6709\u6548\u8bc6\u522b\u548c\u62b5\u5fa1\u5404\u7c7b\u8d8a\u72f1\u653b\u51fb\uff0c\u540c\u65f6\u4fdd\u6301\u6781\u4f4e\u7684\u54cd\u5e94\u5ef6\u8fdf\u3002<\/p>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ba83bf07543a71d36074708f3cc98475.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-ba83bf07543a71d36074708f3cc98475.png\"><\/a><\/section>\n<ul>\n<li>\n<section>\u8bba\u6587\u6807\u9898\uff1aSelfDefend: LLMs Can Defend Themselves against Jailbreaking in a Practical Manner<\/section>\n<\/li>\n<li>\n<section>\u8bba\u6587\u4e3b\u9875\uff1ahttps:\/\/selfdefend.github.io\/\u00a0<\/section>\n<\/li>\n<li>\n<section>\u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/2406.05498<\/section>\n<\/li>\n<li>\n<section>GitHub \u94fe\u63a5\uff1ahttps:\/\/github.com\/selfdefend\/Code<\/section>\n<\/li>\n<\/ul>\n<section><\/section>\n<section>\u8fd1\u5e74\u6765\uff0c\u5927<mark data-type=\"tech_tasks\" data-id=\"fee178e8-ee20-42fc-8f8f-1d41c1f34e8f\">\u8bed\u8a00\u6a21\u578b<\/mark>\uff08LLMs\uff09\u5728<mark data-type=\"tech_tasks\" data-id=\"c8ff5114-6cbb-49ca-8a89-3ee2826be0b4\">\u81ea\u7136\u8bed\u8a00\u5904\u7406<\/mark>\u3001<mark data-type=\"tech_tasks\" data-id=\"197e8428-4002-4317-a18f-993da43b426f\">\u4fe1\u606f\u68c0\u7d22<\/mark>\u3001<mark data-type=\"tech_tasks\" data-id=\"ba1ea8a1-a705-4f89-8348-ec71bae840dc\">\u56fe\u50cf\u751f\u6210<\/mark>\u7b49\u591a\u4e2a\u9886\u57df\u5c55\u73b0\u51fa\u5de8\u5927\u6f5c\u529b\u3002\u7136\u800c\uff0c\u968f\u7740 LLMs \u7684\u5e7f\u6cdb\u5e94\u7528\uff0c\u5982\u4f55\u786e\u4fdd\u5176\u5b89\u5168\u6027\u6210\u4e3a\u4e86\u4e00\u4e2a\u91cd\u8981\u8bfe\u9898\u3002\u5c24\u5176\u662f \u201c\u8d8a\u72f1\u653b\u51fb\u201d\uff08Jailbreaking\uff09\uff0c\u8fd9\u79cd\u653b\u51fb\u901a\u8fc7\u7ed5\u8fc7 LLMs \u7684\u5b89\u5168\u5bf9\u9f50\u673a\u5236\uff0c\u8bf1\u5bfc\u6a21\u578b\u751f\u6210\u6709\u5bb3\u5185\u5bb9\uff0c\u5f15\u53d1\u4e86\u5e7f\u6cdb\u5173\u6ce8\u3002\u4e3a\u4e86\u5e94\u5bf9\u8fd9\u4e00\u6311\u6218\uff0c\u9999\u6e2f\u79d1\u6280\u5927\u5b66\u3001\u5357\u6d0b\u7406\u5de5\u7b49\u56e2\u961f\u8054\u5408\u63d0\u51fa\u4e86\u4e00\u79cd\u540d\u4e3a\u00a0<strong>SelfDefend \u7684\u65b0\u578b\u9632\u5fa1\u6846\u67b6\uff0c<\/strong>\u8be5\u6846\u67b6\u901a\u8fc7\u5f15\u5165 \u201c\u5f71\u5b50 LLM\u201d\uff08Shadow LLM\uff09\u6765\u5e76\u884c\u68c0\u6d4b\u6f5c\u5728\u7684\u6709\u5bb3<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\uff0c\u4ece\u800c\u6709\u6548\u62b5\u5fa1\u591a\u79cd\u8d8a\u72f1\u653b\u51fb\u3002<\/section>\n<section><\/section>\n<section><strong>\u8d8a\u72f1\u653b\u51fb\u7684\u591a\u6837\u6027\u4e0e\u9632\u5fa1\u6311\u6218<\/strong><\/section>\n<section><\/section>\n<section>\u8d8a\u72f1\u653b\u51fb\u7684\u5f62\u5f0f\u591a\u79cd\u591a\u6837\uff0c\u5305\u62ec\u57fa\u4e8e\u4eba\u5de5\u8bbe\u8ba1\u7684\u653b\u51fb\u3001\u57fa\u4e8e\u4f18\u5316\u7684\u653b\u51fb\u3001\u57fa\u4e8e\u751f\u6210\u7684\u653b\u51fb\uff0c\u4ee5\u53ca\u6700\u8fd1\u51fa\u73b0\u7684\u95f4\u63a5\u653b\u51fb\u548c\u591a\u8bed\u8a00\u653b\u51fb\u3002\u8fd9\u4e9b\u653b\u51fb\u624b\u6bb5\u4e0d\u65ad\u8fdb\u5316\uff0c\u4f7f\u5f97\u4f20\u7edf\u7684\u9632\u5fa1\u673a\u5236\u96be\u4ee5\u5e94\u5bf9\u3002\u73b0\u6709\u7684\u9632\u5fa1\u65b9\u6cd5\u4e3b\u8981\u5206\u4e3a\u4e24\u7c7b\uff1a\u57fa\u4e8e\u6a21\u578b\u7684\u9632\u5fa1\u548c\u57fa\u4e8e\u63d2\u4ef6\u7684\u9632\u5fa1\u3002\u524d\u8005\u901a\u8fc7\u6539\u8fdb\u6a21\u578b\u7684\u5185\u5728\u673a\u5236\u6765\u589e\u5f3a\u5b89\u5168\u6027\uff0c\u540e\u8005\u5219\u901a\u8fc7\u5916\u90e8\u63d2\u4ef6\u6765\u589e\u5f3a\u73b0\u6709\u6a21\u578b\u7684\u5b89\u5168\u6027\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u65b9\u6cd5\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u9762\u4e34\u8bf8\u591a\u6311\u6218\uff0c\u65e0\u6cd5\u540c\u65f6\u6ee1\u8db3\u56db\u4e2a\u76ee\u6807\uff1a\u5e94\u5bf9\u6240\u6709\u7c7b\u578b\u7684\u653b\u51fb\uff08O1\uff09\u3001\u5f15\u5165\u53ef\u5ffd\u7565\u7684\u989d\u5916\u5ef6\u8fdf\uff08O2\uff09\u3001\u5bf9\u68c0\u6d4b\u51fa\u7684\u8d8a\u72f1\u8bbf\u95ee\u63d0\u4f9b\u53ef\u89e3\u91ca\u6027\uff08O3\uff09\uff0c\u4ee5\u53ca\u540c\u65f6\u9002\u7528\u4e8e\u5f00\u6e90\u548c\u95ed\u6e90\u6a21\u578b\uff08O4\uff09\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-260484cf50b834cd7aa44bf084e8ca50.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-260484cf50b834cd7aa44bf084e8ca50.png\"><\/a><\/section>\n<section><strong>SelfDefend \u6846\u67b6\u7684\u521b\u65b0\u8bbe\u8ba1<\/strong><\/section>\n<section><\/section>\n<section>SelfDefend \u6846\u67b6\u7684\u7075\u611f\u6765\u6e90\u4e8e\u4f20\u7edf\u5b89\u5168\u9886\u57df\u4e2d\u7684 \u201c\u5f71\u5b50\u6808\u201d\uff08Shadow Stack\uff09\u6982\u5ff5\u3002\u5f71\u5b50\u6808\u901a\u8fc7\u521b\u5efa\u4e00\u4e2a\u5e76\u884c\u7684\u5185\u5b58\u7a7a\u95f4\u6765\u9632\u5fa1\u5185\u5b58\u6ea2\u51fa\u653b\u51fb\uff0c\u800c SelfDefend \u5219\u901a\u8fc7\u521b\u5efa\u4e00\u4e2a\u5e76\u884c\u7684 \u201c\u5f71\u5b50 LLM\u201d \u6765\u68c0\u6d4b\u6f5c\u5728\u7684\u6709\u5bb3<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\u3002\u5177\u4f53\u6765\u8bf4\uff0cSelfDefend \u6846\u67b6\u5305\u542b\u4e24\u4e2a\u5e76\u884c\u7684 LLM \u5b9e\u4f8b\uff1a\u4e00\u4e2a\u7528\u4e8e\u6b63\u5e38\u54cd\u5e94\u7528\u6237<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\u7684<strong>\u76ee\u6807 LLM<\/strong>\uff08<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-a20662946a13233315291150a6713636.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-a20662946a13233315291150a6713636.png\"><\/a>\uff09\uff0c\u53e6\u4e00\u4e2a\u7528\u4e8e\u68c0\u6d4b\u6709\u5bb3\u5185\u5bb9\u7684<strong>\u9632\u5fa1 LLM<\/strong>\uff08<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5681b0ba75f8cb06af22887e3b40512a.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-5681b0ba75f8cb06af22887e3b40512a.png\"><\/a>\uff09\u3002\u5f53\u7528\u6237\u8f93\u5165<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\u65f6\uff0c\u76ee\u6807 LLM \u4f1a\u6b63\u5e38\u5904\u7406<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\u5e76\u751f\u6210\u54cd\u5e94\uff0c\u800c\u9632\u5fa1 LLM \u5219\u901a\u8fc7\u7279\u5b9a\u7684\u68c0\u6d4b\u63d0\u793a\u8bcd\uff08<a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-65fc6794490e76e28d33c33533cf8960.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-65fc6794490e76e28d33c33533cf8960.png\"><\/a>\uff09\u6765\u8bc6\u522b<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\u4e2d\u7684\u6709\u5bb3\u90e8\u5206\u6216\u610f\u56fe\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-90bd8ccbe7c3e48e89f2c16822021e71.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-90bd8ccbe7c3e48e89f2c16822021e71.png\"><\/a><\/section>\n<section>\u8fd9\u79cd\u8bbe\u8ba1\u5e26\u6765\u4e86\u591a\u91cd\u4f18\u52bf\uff1a\u9996\u5148\uff0c\u5b83\u540c\u65f6\u5229\u7528\u4e86\u76ee\u6807 LLM \u7684\u5b89\u5168\u5bf9\u9f50\u673a\u5236\u548c\u9632\u5fa1 LLM \u7684\u8d8a\u72f1\u68c0\u6d4b\u80fd\u529b\uff0c\u5f62\u6210\u4e86\u53cc\u91cd\u4fdd\u62a4\u5c42\uff0c\u663e\u8457\u63d0\u9ad8\u4e86\u9632\u5fa1\u6210\u529f\u7387\uff1b\u5176\u6b21\uff0c\u7531\u4e8e\u9632\u5fa1 LLM \u7684\u8f93\u51fa\u901a\u5e38\u8f83\u77ed\uff08\u5982 \u201cNo\u201d \u8868\u793a\u65e0\u95ee\u9898\uff09\uff0c\u6b63\u5e38<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\u7684\u54cd\u5e94\u5ef6\u8fdf\u51e0\u4e4e\u53ef\u4ee5\u5ffd\u7565\u4e0d\u8ba1\uff1b\u7136\u540e\u68c0\u6d4b\u51fa\u7684\u6709\u5bb3\u90e8\u5206\u6216\u8005\u6076\u610f\u610f\u56fe\u53ef\u4ee5\u4f5c\u4e3a\u9632\u5fa1\u7684\u53ef\u89e3\u91ca\u6027\uff1b\u6700\u540e\uff0c\u7531\u4e8e\u9632\u5fa1 LLM \u4e0d\u9700\u8981\u4fee\u6539\u6216\u76d1\u63a7\u76ee\u6807 LLM \u7684\u5185\u90e8\u673a\u5236\uff0c\u56e0\u6b64\u53ef\u4ee5\u517c\u5bb9\u5f00\u6e90\u548c\u95ed\u6e90\u6a21\u578b\u3002<\/section>\n<section><\/section>\n<section><strong>\u5b9e\u9a8c\u9a8c\u8bc1\u4e0e\u6548\u679c\u8bc4\u4f30<\/strong><\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-3b79e6d1f16caff7480666fe4a1613dc.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-3b79e6d1f16caff7480666fe4a1613dc.png\"><\/a><\/section>\n<section>\u7814\u7a76\u56e2\u961f\u901a\u8fc7\u5927\u91cf\u5b9e\u9a8c\u9a8c\u8bc1\u4e86 SelfDefend \u6846\u67b6\u7684\u6709\u6548\u6027\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u57fa\u4e8e GPT-3.5 \u548c GPT-4 \u7684 SelfDefend \u80fd\u591f\u663e\u8457\u964d\u4f4e\u591a\u79cd\u8d8a\u72f1\u653b\u51fb\u7684\u6210\u529f\u7387\u3002\u4f8b\u5982\uff0c\u57fa\u4e8e GPT-3.5 \u7684 SelfDefend \u5c06\u653b\u51fb\u6210\u529f\u7387\uff08ASR\uff09\u4ece\u5e73\u5747 65.7% \u964d\u4f4e\u81f3 0.236\uff0c\u800c\u57fa\u4e8e GPT-4 \u7684 SelfDefend \u66f4\u662f\u5c06 ASR \u964d\u4f4e\u81f3\u5e73\u5747 0.050\u3002\u6b64\u5916\uff0cSelfDefend \u5bf9\u6b63\u5e38<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\u7684\u5f71\u54cd\u5fae\u4e4e\u5176\u5fae\uff0cGPT-3.5 \u548c GPT-4 \u7684\u6b63\u5e38<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\u901a\u8fc7\u7387\u4ec5\u5206\u522b\u4e0b\u964d\u4e86 0.51% \u548c 2.77%\u3002<\/section>\n<section><\/section>\n<section>\u4e3a\u4e86\u8fdb\u4e00\u6b65\u964d\u4f4e\u6210\u672c\u548c\u63d0\u5347\u9c81\u68d2\u6027\uff0c\u7814\u7a76\u56e2\u961f\u8fd8\u901a\u8fc7\u6570\u636e\u84b8\u998f\u65b9\u6cd5\u5bf9\u5f00\u6e90\u7684 Llama-2-7b \u6a21\u578b\u8fdb\u884c\u4e86\u5fae\u8c03\uff0c\u751f\u6210\u4e86\u4e13\u7528\u7684\u9632\u5fa1\u6a21\u578b\u3002\u5b9e\u9a8c\u8868\u660e\uff0c\u8fd9\u4e9b\u5fae\u8c03\u540e\u7684\u6a21\u578b\u5728\u9632\u5fa1\u6548\u679c\u4e0a\u4e0e\u57fa\u4e8e GPT-4 \u7684 SelfDefend \u76f8\u5f53\uff0c\u4e14\u989d\u5916\u5ef6\u8fdf\u663e\u8457\u964d\u4f4e\u3002\u4f8b\u5982\uff0c\u5fae\u8c03\u540e\u7684\u6a21\u578b\u5728\u6b63\u5e38<mark data-type=\"concepts\" data-id=\"bf740558-f0f7-41a8-87a0-e695a97563b3\">\u67e5\u8be2<\/mark>\u4e2d\u7684\u5e73\u5747\u5ef6\u8fdf\u4ec5\u4e3a 0-0.01 \u79d2\uff0c\u800c\u5728\u653b\u51fb\u573a\u666f\u4e2d\u7684\u6700\u5927\u5ef6\u8fdf\u4ece GPT-4 \u7684 1.56 \u79d2\u964d\u4f4e\u81f3 0.39 \u79d2\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c9c06380920b739eae57378f061c78a8.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-c9c06380920b739eae57378f061c78a8.png\"><\/a><\/section>\n<section><strong>\u4e0e\u73b0\u6709\u9632\u5fa1\u65b9\u6cd5\u7684\u5bf9\u6bd4<\/strong><\/section>\n<section><strong><br \/><\/strong><\/section>\n<section>\u7814\u7a76\u56e2\u961f\u8fd8\u5c06 SelfDefend \u4e0e\u73b0\u6709\u7684\u4e03\u79cd\u4e3b\u6d41\u9632\u5fa1\u65b9\u6cd5\u8fdb\u884c\u4e86\u5bf9\u6bd4\uff0c\u5305\u62ec ICD\u3001SafeDecoding\u3001Perplexity Filter\u3001SmoothLLM\u3001Llama Guard \u7b49\u3002\u5b9e\u9a8c\u7ed3\u679c\u663e\u793a\uff0cSelfDefend \u5728 60 \u4e2a\u6d4b\u8bd5\u573a\u666f\u4e2d\u7684 55 \u4e2a\u573a\u666f\u4e2d\u8868\u73b0\u6700\u4f18\uff0c\u5c24\u5176\u662f\u5728\u5e94\u5bf9\u95f4\u63a5\u653b\u51fb\u548c\u591a\u8bed\u8a00\u653b\u51fb\u65f6\uff0cSelfDefend \u7684\u9632\u5fa1\u6548\u679c\u663e\u8457\u4f18\u4e8e\u5176\u4ed6\u65b9\u6cd5\u3002\u6b64\u5916\uff0cSelfDefend \u7684\u989d\u5916\u5ef6\u8fdf\u4e5f\u8fdc\u4f4e\u4e8e\u5176\u4ed6\u9632\u5fa1\u65b9\u6cd5\uff0c\u4f7f\u5176\u5728\u5b9e\u9645\u90e8\u7f72\u4e2d\u66f4\u5177\u53ef\u884c\u6027\u3002<\/section>\n<section><a href=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-37c3a6e8794fbd7426aa9079e0cf20b9.png\" data-fancybox=\"images\" data-fancybox=\"gallery\"><img decoding=\"async\" src=\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-37c3a6e8794fbd7426aa9079e0cf20b9.png\"><\/a><\/section>\n<section><strong>\u672a\u6765\u5c55\u671b<\/strong><\/section>\n<section><\/section>\n<section>\u8fd9\u9879\u5f00\u521b\u6027\u7684\u7814\u7a76\u4e0d\u4ec5\u4e3a AI \u5b89\u5168\u9886\u57df\u5e26\u6765\u4e86\u7a81\u7834\u6027\u8fdb\u5c55\uff0c\u66f4\u63ed\u793a\u4e86\u4e00\u4e2a\u632f\u594b\u4eba\u5fc3\u7684\u4fe1\u53f7\uff1aAI \u7cfb\u7edf\u7684\u5b89\u5168\u6027\u4e0e\u6548\u7387\u4e0d\u518d\u662f\u9c7c\u548c\u718a\u638c\u4e0d\u53ef\u517c\u5f97\u3002\u901a\u8fc7\u8d4b\u4e88 AI&#8217; \u81ea\u536b\u610f\u8bc6 &#8216;\uff0cSelfDefend \u5c55\u73b0\u4e86\u4e00\u4e2a\u66f4\u5b89\u5168\u7684 AI \u672a\u6765\uff1a\u5728\u8fd9\u4e2a\u672a\u6765\u91cc\uff0cAI \u7cfb\u7edf\u65e2\u80fd\u4fdd\u6301\u9ad8\u6548\u670d\u52a1\u80fd\u529b\uff0c\u53c8\u80fd\u4e3b\u52a8\u8bc6\u522b\u548c\u62b5\u5fa1\u6f5c\u5728\u5a01\u80c1\uff0c\u771f\u6b63\u5b9e\u73b0 &#8216; \u81ea\u6211\u5b88\u62a4 &#8216;\u3002<\/section>\n<section><\/section>\n<section><em><sup>\u53c2\u8003\u94fe\u63a5\uff1a<\/sup><\/em><\/section>\n<section><sup><em>https:\/\/x.com\/rohanpaul_ai\/status\/1886025249273339961<\/em><\/sup><\/section>\n<section><em><sup>https:\/\/techcrunch.com\/2025\/02\/09\/deepseeks-r1-reportedly-more-vulnerable-to-jailbreaking-than-other-ai-models\/<\/sup><\/em><\/section>\n<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:<a href=\"https:\/\/www.jiqizhixin.com\/articles\/2025-02-11-4\" target=\"_blank\">\u7f51\u4f20DeepSeek R1\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\uff1f\u8fd9\u6709\u4e2a\u5165\u9009\u9876\u4f1a\u7684\u9632\u5fa1\u6846\u67b6SelfDefend<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6587\u7ae0\u6765\u6e90\u4e8e\u4e92\u8054\u7f51:\u7f51\u4f20DeepSeek  [&hellip;]<\/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-38237","post","type-post","status-publish","format-standard","hentry","category-news","tag-68","tag-67"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.4 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>\u7f51\u4f20DeepSeek R1\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\uff1f\u8fd9\u6709\u4e2a\u5165\u9009\u9876\u4f1a\u7684\u9632\u5fa1\u6846\u67b6SelfDefend - \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=38237\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/17aitech.com\/?p=38237\",\"url\":\"https:\/\/17aitech.com\/?p=38237\",\"name\":\"\u7f51\u4f20DeepSeek R1\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\uff1f\u8fd9\u6709\u4e2a\u5165\u9009\u9876\u4f1a\u7684\u9632\u5fa1\u6846\u67b6SelfDefend - \u4e00\u8d77AI\u6280\u672f\",\"isPartOf\":{\"@id\":\"https:\/\/17aitech.com\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/17aitech.com\/?p=38237#primaryimage\"},\"image\":{\"@id\":\"https:\/\/17aitech.com\/?p=38237#primaryimage\"},\"thumbnailUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f9a9ecb29e93345299a082069b54dbc5.png\",\"datePublished\":\"2025-02-12T03:00:55+00:00\",\"author\":{\"@id\":\"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d\"},\"breadcrumb\":{\"@id\":\"https:\/\/17aitech.com\/?p=38237#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/17aitech.com\/?p=38237\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/17aitech.com\/?p=38237#primaryimage\",\"url\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f9a9ecb29e93345299a082069b54dbc5.png\",\"contentUrl\":\"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f9a9ecb29e93345299a082069b54dbc5.png\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/17aitech.com\/?p=38237#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/17aitech.com\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u7f51\u4f20DeepSeek R1\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\uff1f\u8fd9\u6709\u4e2a\u5165\u9009\u9876\u4f1a\u7684\u9632\u5fa1\u6846\u67b6SelfDefend\"}]},{\"@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":"\u7f51\u4f20DeepSeek R1\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\uff1f\u8fd9\u6709\u4e2a\u5165\u9009\u9876\u4f1a\u7684\u9632\u5fa1\u6846\u67b6SelfDefend - \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=38237","schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/17aitech.com\/?p=38237","url":"https:\/\/17aitech.com\/?p=38237","name":"\u7f51\u4f20DeepSeek R1\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\uff1f\u8fd9\u6709\u4e2a\u5165\u9009\u9876\u4f1a\u7684\u9632\u5fa1\u6846\u67b6SelfDefend - \u4e00\u8d77AI\u6280\u672f","isPartOf":{"@id":"https:\/\/17aitech.com\/#website"},"primaryImageOfPage":{"@id":"https:\/\/17aitech.com\/?p=38237#primaryimage"},"image":{"@id":"https:\/\/17aitech.com\/?p=38237#primaryimage"},"thumbnailUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f9a9ecb29e93345299a082069b54dbc5.png","datePublished":"2025-02-12T03:00:55+00:00","author":{"@id":"https:\/\/17aitech.com\/#\/schema\/person\/60225458499e817ae0af73e67e440b9d"},"breadcrumb":{"@id":"https:\/\/17aitech.com\/?p=38237#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/17aitech.com\/?p=38237"]}]},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/17aitech.com\/?p=38237#primaryimage","url":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f9a9ecb29e93345299a082069b54dbc5.png","contentUrl":"https:\/\/17aitech.com\/wp-content\/uploads\/2025\/02\/frc-f9a9ecb29e93345299a082069b54dbc5.png"},{"@type":"BreadcrumbList","@id":"https:\/\/17aitech.com\/?p=38237#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/17aitech.com\/"},{"@type":"ListItem","position":2,"name":"\u7f51\u4f20DeepSeek R1\u66f4\u5bb9\u6613\u88ab\u8d8a\u72f1\uff1f\u8fd9\u6709\u4e2a\u5165\u9009\u9876\u4f1a\u7684\u9632\u5fa1\u6846\u67b6SelfDefend"}]},{"@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\/38237","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=38237"}],"version-history":[{"count":0,"href":"https:\/\/17aitech.com\/index.php?rest_route=\/wp\/v2\/posts\/38237\/revisions"}],"wp:attachment":[{"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=38237"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=38237"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/17aitech.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=38237"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}