{"id":186,"date":"2023-02-09T17:50:32","date_gmt":"2023-02-09T09:50:32","guid":{"rendered":"http:\/\/www.bosombot.com\/?p=186"},"modified":"2023-02-12T12:02:21","modified_gmt":"2023-02-12T04:02:21","slug":"chatgpt%e5%8f%91%e5%b1%95%e5%8e%86%e7%a8%8b%e3%80%81%e5%8e%9f%e7%90%86%e3%80%81%e6%8a%80%e6%9c%af%e6%9e%b6%e6%9e%84%e8%af%a6%e8%a7%a3%e5%92%8c%e4%ba%a7%e4%b8%9a%e6%9c%aa%e6%9d%a5","status":"publish","type":"post","link":"https:\/\/www.bosombot.com\/?p=186","title":{"rendered":"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765"},"content":{"rendered":"\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-8f761849 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\" style=\"flex-basis:100%\">\n<p class=\"wp-block-paragraph\"><strong>\u6765\u6e90\uff1a\u9648\u5dcd\u8c08\u82af\uff0c<\/strong>\u672c\u6587\u5c06\u4ecb\u7ecdChatGPT\u7684<strong>\u7279\u70b9\u3001\u529f\u80fd\u3001\u6280\u672f\u67b6\u6784\u3001\u5c40\u9650\u3001\u4ea7\u4e1a\u5e94\u7528\u3001\u6295\u8d44\u673a\u4f1a\u548c\u672a\u6765<\/strong>\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u539f\u6587\u94fe\u63a5\uff1a<a href=\"https:\/\/zhuanlan.zhihu.com\/p\/590655677\">https:\/\/zhuanlan.zhihu.com\/p\/590655677<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4f5c\u8005\uff1a\u9648\u5dcd \u535a\u58eb\uff0c\u4f5c\u8005\u672c\u4eba\u66fe\u62c5\u4efb\u534e\u4e3a\u7cfb\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08 NLP \uff09\u4f01\u4e1a\u7684\u9996\u5e2d\u79d1\u5b66\u5bb6\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">0\uff0c\u5f15\u8a00<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u5148\u4e0a\u53c2\u8003\u7f51\u9875\u6216\u8bba\u6587\u3002\u4e13\u4e1a\u7684\u8bfb\u8005\u53ef\u4ee5\u76f4\u63a5\u770bpaper\u3002<\/p>\n\n\n\n<div class=\"wp-block-columns is-layout-flex wp-container-core-columns-is-layout-8f761849 wp-block-columns-is-layout-flex\">\n<div class=\"wp-block-column is-layout-flow wp-block-column-is-layout-flow\">\n<ol class=\"wp-block-list\">\n<li>ChatGPT: Optimizing Language Models for Dialogue&nbsp;ChatGPT: Optimizing Language Models for Dialogue<\/li>\n\n\n\n<li>GPT\u8bba\u6587\uff1aLanguage Models are Few-Shot Learners Language Models are Few-Shot Learners<\/li>\n\n\n\n<li>InstructGPT\u8bba\u6587\uff1aTraining language models to follow instructions with human feedback Training language models to follow instructions with human feedback<\/li>\n\n\n\n<li>huggingface\u89e3\u8bfbRHLF\u7b97\u6cd5\uff1aIllustrating Reinforcement Learning from Human Feedback (RLHF) Illustrating Reinforcement Learning from Human Feedback (RLHF)<\/li>\n\n\n\n<li>RHLF\u7b97\u6cd5\u8bba\u6587\uff1aAugmenting Reinforcement Learning with Human Feedback cs.utexas.edu\/~ai-lab\/p<\/li>\n\n\n\n<li>TAMER\u6846\u67b6\u8bba\u6587\uff1aInteractively Shaping Agents via Human Reinforcement cs.utexas.edu\/~bradknox<\/li>\n\n\n\n<li>PPO\u7b97\u6cd5\uff1aProximal Policy Optimization Algorithms Proximal Policy Optimization Algorithms<\/li>\n<\/ol>\n<\/div>\n<\/div>\n\n\n\n<p class=\"wp-block-paragraph\">\u4eca\u5e7412\u67081\u65e5\uff0cOpenAI\u63a8\u51fa\u4eba\u5de5\u667a\u80fd\u804a\u5929\u539f\u578b<strong>ChatGPT<\/strong>\uff0c\u518d\u6b21\u8d5a\u8db3\u773c\u7403\uff0c\u4e3aAI\u754c\u5f15\u53d1\u4e86\u7c7b\u4f3cAIGC\u8ba9\u827a\u672f\u5bb6\u5931\u4e1a\u7684\u5927\u8ba8\u8bba\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u636e\u62a5\u9053\uff0cChatGPT\u5728\u5f00\u653e\u8bd5\u7528\u7684\u77ed\u77ed\u51e0\u5929\uff0c\u5c31\u5438\u5f15\u4e86\u8d85\u8fc7<strong>&nbsp;100 \u4e07\u4e92\u8054\u7f51\u6ce8\u518c\u7528\u6237<\/strong>\u3002\u5e76\u4e14\u793e\u4ea4\u7f51\u7edc\u6d41\u4f20\u51fa\u5404\u79cd\u8be2\u95ee\u6216\u8c03\u620fChatGPT\u7684<strong>\u6709\u8da3\u5bf9\u8bdd<\/strong>\u3002\u751a\u81f3\u6709\u4eba\u5c06ChatGPT\u6bd4\u55bb\u4e3a\u201c<strong>\u641c\u7d22\u5f15\u64ce+\u793e\u4ea4\u8f6f\u4ef6<\/strong>\u201d\u7684\u7ed3\u5408\u4f53\uff0c\u80fd\u591f\u5728\u5b9e\u65f6\u4e92\u52a8\u7684\u8fc7\u7a0b\u4e2d\u83b7\u5f97\u95ee\u9898\u7684\u5408\u7406\u7b54\u6848\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ChatGPT \u662f\u4e00\u79cd\u4e13\u6ce8\u4e8e\u5bf9\u8bdd\u751f\u6210\u7684<strong>\u8bed\u8a00\u6a21\u578b<\/strong>\u3002\u5b83\u80fd\u591f\u6839\u636e\u7528\u6237\u7684\u6587\u672c\u8f93\u5165\uff0c\u4ea7\u751f\u76f8\u5e94\u7684\u667a\u80fd\u56de\u7b54\u3002\u8fd9\u4e2a\u56de\u7b54\u53ef\u4ee5\u662f\u7b80\u77ed\u7684\u8bcd\u8bed\uff0c\u4e5f\u53ef\u4ee5\u662f<strong>\u957f\u7bc7\u5927\u8bba<\/strong>\u3002\u5176\u4e2dGPT\u662fGenerative Pre-trained Transformer\uff08\u751f\u6210\u578b\u9884\u8bad\u7ec3\u53d8\u6362\u6a21\u578b\uff09\u7684\u7f29\u5199\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u901a\u8fc7\u5b66\u4e60\u5927\u91cf\u73b0\u6210\u6587\u672c\u548c\u5bf9\u8bdd\u96c6\u5408\uff08\u4f8b\u5982Wiki\uff09\uff0cChatGPT\u80fd\u591f\u50cf\u4eba\u7c7b\u90a3\u6837\u5373\u65f6\u5bf9\u8bdd\uff0c\u6d41\u7545\u7684\u56de\u7b54\u5404\u79cd\u95ee\u9898\u3002\uff08\u5f53\u7136\u56de\u7b54\u901f\u5ea6\u6bd4\u4eba\u8fd8\u662f\u6162\u4e00\u4e9b\uff09\u65e0\u8bba\u662f\u82f1\u6587\u8fd8\u662f\u5176\u4ed6\u8bed\u8a00\uff08\u4f8b\u5982\u4e2d\u6587\u3001\u97e9\u8bed\u7b49\uff09\uff0c\u4ece\u56de\u7b54\u5386\u53f2\u95ee\u9898\uff0c\u5230\u5199\u6545\u4e8b\uff0c\u751a\u81f3\u662f<strong>\u64b0\u5199\u5546\u4e1a\u8ba1\u5212\u4e66\u548c\u884c\u4e1a\u5206\u6790<\/strong>\uff0c\u201c\u51e0\u4e4e\u201d\u65e0\u6240\u4e0d\u80fd\u3002\u751a\u81f3\u6709\u7a0b\u5e8f\u5458\u8d34\u51fa\u4e86ChatGPT\u8fdb\u884c\u7a0b\u5e8f\u4fee\u6539\u7684\u5bf9\u8bdd\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter\"><img decoding=\"async\" src=\"https:\/\/pic2.zhimg.com\/80\/v2-cda1af4d2f17bd11dce94cafd580dd61_720w.webp\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe\" \/><figcaption class=\"wp-element-caption\">ChatGPT\u548cAIGC\u7684\u8054\u5408\u4f7f\u7528<\/figcaption><\/figure>\n<\/div>\n\n\n<p class=\"wp-block-paragraph\">ChatGPT\u4e5f\u53ef\u4ee5\u4e0e\u5176\u4ed6AIGC\u6a21\u578b<strong>\u8054\u5408\u4f7f\u7528<\/strong>\uff0c\u83b7\u5f97\u66f4\u52a0\u70ab\u9177\u5b9e\u7528\u7684\u529f\u80fd\u3002\u4f8b\u5982\u4e0a\u9762\u901a\u8fc7\u5bf9\u8bdd\u751f\u6210\u5ba2\u5385\u8bbe\u8ba1\u56fe\u3002\u8fd9\u6781\u5927\u52a0\u5f3a\u4e86AI\u5e94\u7528\u4e0e\u5ba2\u6237\u5bf9\u8bdd\u7684\u80fd\u529b\uff0c\u4f7f\u6211\u4eec\u770b\u5230\u4e86<strong>AI\u5927\u89c4\u6a21\u843d\u5730\u7684\u66d9\u5149<\/strong>\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">1\uff0cChatGPT\u7684\u4f20\u627f\u4e0e\u7279\u70b9<\/h2>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"400\" height=\"76\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image.png\"  class=\"wp-image-188\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe1\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe1\" \/><\/figure>\n<\/div>\n\n\n<h3 class=\"wp-block-heading\">1.1 OpenAI\u5bb6\u65cf<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u6211\u4eec\u9996\u5148\u4e86\u89e3\u4e0b<strong>OpenAI<\/strong>\u662f\u54ea\u8def\u5927\u795e\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">OpenAI\u603b\u90e8\u4f4d\u4e8e\u65e7\u91d1\u5c71\uff0c\u7531\u7279\u65af\u62c9\u7684\u9a6c\u65af\u514b\u3001Sam Altman\u53ca\u5176\u4ed6\u6295\u8d44\u8005\u57282015\u5e74\u5171\u540c\u521b\u7acb\uff0c\u76ee\u6807\u662f\u5f00\u53d1\u9020\u798f\u5168\u4eba\u7c7b\u7684AI\u6280\u672f\u3002\u800c\u9a6c\u65af\u514b\u5219\u57282018\u5e74\u65f6\u56e0\u516c\u53f8\u53d1\u5c55\u65b9\u5411\u5206\u6b67\u800c\u79bb\u5f00\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6b64\u524d\uff0cOpenAI \u56e0\u63a8\u51fa&nbsp;<strong>GPT\u7cfb\u5217\u81ea\u7136\u8bed\u8a00\u5904\u7406\u6a21\u578b<\/strong>\u800c\u95fb\u540d\u3002\u4ece2018\u5e74\u8d77\uff0cOpenAI\u5c31\u5f00\u59cb\u53d1\u5e03\u751f\u6210\u5f0f\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578bGPT\uff08Generative Pre-trained Transformer\uff09\uff0c\u53ef\u7528\u4e8e\u751f\u6210\u6587\u7ae0\u3001\u4ee3\u7801\u3001\u673a\u5668\u7ffb\u8bd1\u3001\u95ee\u7b54\u7b49\u5404\u7c7b\u5185\u5bb9\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6bcf\u4e00\u4ee3GPT\u6a21\u578b\u7684\u53c2\u6570\u91cf\u90fd<strong>\u7206\u70b8\u5f0f\u589e\u957f<\/strong>\uff0c\u582a\u79f0\u201c\u8d8a\u5927\u8d8a\u597d\u201d\u30022019\u5e742\u6708\u53d1\u5e03\u7684GPT-2\u53c2\u6570\u91cf\u4e3a15\u4ebf\uff0c\u800c2020\u5e745\u6708\u7684GPT-3\uff0c\u53c2\u6570\u91cf\u8fbe\u5230\u4e861750\u4ebf\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"1083\" height=\"193\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-16.png\"  class=\"wp-image-216\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe2\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe2\" \/><figcaption class=\"wp-element-caption\">GPT\u5bb6\u65cf\u4e3b\u8981\u6a21\u578b\u5bf9\u6bd4<\/figcaption><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">1.2 ChatGPT\u7684\u4e3b\u8981\u7279\u70b9<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>ChatGPT&nbsp;<\/strong>\u662f\u57fa\u4e8eGPT-3.5\uff08Generative Pre-trained Transformer 3.5\uff09\u67b6\u6784\u5f00\u53d1\u7684\u5bf9\u8bddAI\u6a21\u578b\uff0c\u662fInstructGPT \u7684\u5144\u5f1f\u6a21\u578b\u3002ChatGPT\u5f88\u53ef\u80fd\u662fOpenAI \u5728GPT-4 \u6b63\u5f0f\u63a8\u51fa\u4e4b\u524d\u7684\u6f14\u7ec3\uff0c\u6216\u7528\u4e8e<strong>\u6536\u96c6\u5927\u91cf\u5bf9\u8bdd\u6570\u636e\u3002<\/strong><\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"626\" height=\"418\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-14.png\"  class=\"wp-image-202\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe3\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe3\" \/><figcaption class=\"wp-element-caption\">ChatGPT\u7684\u4e3b\u8981\u7279\u70b9<\/figcaption><\/figure>\n<\/div>\n\n\n<p class=\"wp-block-paragraph\">OpenAI\u4f7f\u7528&nbsp;<strong>RLHF<\/strong>\uff08Reinforcement Learning from Human Feedbac\uff0c\u4eba\u7c7b\u53cd\u9988\u5f3a\u5316\u5b66\u4e60\uff09 \u6280\u672f\u5bf9 ChatGPT \u8fdb\u884c\u4e86\u8bad\u7ec3\uff0c\u4e14\u52a0\u5165\u4e86\u66f4\u591a<strong>\u4eba\u5de5\u76d1\u7763<\/strong>\u8fdb\u884c\u5fae\u8c03\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6b64\u5916\uff0cChatGPT \u8fd8\u5177\u6709\u4ee5\u4e0b\u7279\u5f81\uff1a<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">1\uff09\u53ef\u4ee5\u4e3b\u52a8\u627f\u8ba4\u81ea\u8eab\u9519\u8bef\u3002\u82e5\u7528\u6237\u6307\u51fa\u5176\u9519\u8bef\uff0c\u6a21\u578b\u4f1a\u542c\u53d6\u610f\u89c1\u5e76\u4f18\u5316\u7b54\u6848\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">2\uff09ChatGPT \u53ef\u4ee5\u8d28\u7591\u4e0d\u6b63\u786e\u7684\u95ee\u9898\u3002\u4f8b\u5982\u88ab\u8be2\u95ee \u201c\u54e5\u4f26\u5e03 2015 \u5e74\u6765\u5230\u7f8e\u56fd\u7684\u60c5\u666f\u201d \u7684\u95ee\u9898\u65f6\uff0c\u673a\u5668\u4eba\u4f1a\u8bf4\u660e\u54e5\u4f26\u5e03\u4e0d\u5c5e\u4e8e\u8fd9\u4e00\u65f6\u4ee3\u5e76\u8c03\u6574\u8f93\u51fa\u7ed3\u679c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">3\uff09ChatGPT \u53ef\u4ee5\u627f\u8ba4\u81ea\u8eab\u7684\u65e0\u77e5\uff0c\u627f\u8ba4\u5bf9\u4e13\u4e1a\u6280\u672f\u7684\u4e0d\u4e86\u89e3\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">4\uff09\u652f\u6301\u8fde\u7eed\u591a\u8f6e\u5bf9\u8bdd\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e0e\u5927\u5bb6\u5728\u751f\u6d3b\u4e2d\u7528\u5230\u7684\u5404\u7c7b\u667a\u80fd\u97f3\u7bb1\u548c\u201c<strong>\u4eba\u5de5\u667a\u969c<\/strong>\u201c\u4e0d\u540c\uff0cChatGPT\u5728\u5bf9\u8bdd\u8fc7\u7a0b\u4e2d\u4f1a\u8bb0\u5fc6\u5148\u524d\u4f7f\u7528\u8005\u7684\u5bf9\u8bdd\u8baf\u606f\uff0c\u5373\u4e0a\u4e0b\u6587\u7406\u89e3\uff0c\u4ee5\u56de\u7b54\u67d0\u4e9b\u5047\u8bbe\u6027\u7684\u95ee\u9898\u3002ChatGPT\u53ef\u5b9e\u73b0<strong>\u8fde\u7eed\u5bf9\u8bdd<\/strong>\uff0c\u6781\u5927\u7684\u63d0\u5347\u4e86<strong>\u5bf9\u8bdd\u4ea4\u4e92\u6a21\u5f0f<\/strong>\u4e0b\u7684\u7528\u6237\u4f53\u9a8c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5bf9\u4e8e\u51c6\u786e\u7ffb\u8bd1\u6765\u8bf4\uff08\u5c24\u5176\u662f\u4e2d\u6587\u4e0e\u4eba\u540d\u97f3\u8bd1\uff09\uff0cChatGPT\u79bb\u5b8c\u7f8e\u8fd8\u6709\u4e00\u6bb5\u8ddd\u79bb\uff0c\u4e0d\u8fc7\u5728\u6587\u5b57\u6d41\u7545\u5ea6\u4ee5\u53ca\u8fa8\u522b\u7279\u5b9a\u4eba\u540d\u6765\u8bf4\uff0c\u4e0e\u5176\u4ed6\u7f51\u7edc\u7ffb\u8bd1\u5de5\u5177\u76f8\u8fd1\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7531\u4e8e ChatGPT\u662f\u4e00\u4e2a\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff0c\u76ee\u524d\u8fd8\u5e76<strong>\u4e0d\u5177\u5907\u7f51\u7edc\u641c\u7d22\u529f\u80fd<\/strong>\uff0c\u56e0\u6b64\u5b83\u53ea\u80fd\u57fa\u4e8e2021\u5e74\u6240\u62e5\u6709\u7684\u6570\u636e\u96c6\u8fdb\u884c\u56de\u7b54\u3002\u4f8b\u5982\u5b83\u4e0d\u77e5\u90532022\u5e74\u4e16\u754c\u676f\u7684\u60c5\u51b5\uff0c\u4e5f\u4e0d\u4f1a\u50cf\u82f9\u679c\u7684Siri\u90a3\u6837\u56de\u7b54\u4eca\u5929\u5929\u6c14\u5982\u4f55\u3001\u6216\u5e2e\u4f60\u641c\u7d22\u4fe1\u606f\u3002\u5982\u679cChatGPT\u80fd\u4e0a\u7f51\u81ea\u5df1\u5bfb\u627e\u5b66\u4e60\u8bed\u6599\u548c\u641c\u7d22\u77e5\u8bc6\uff0c\u4f30\u8ba1\u53c8\u4f1a\u6709\u66f4\u5927\u7684\u7a81\u7834\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5373\u4fbf\u5b66\u4e60\u7684\u77e5\u8bc6\u6709\u9650\uff0cChatGPT \u8fd8\u662f\u80fd\u56de\u7b54\u8111\u6d1e\u5927\u5f00\u7684\u4eba\u7c7b\u7684\u8bb8\u591a<strong>\u5947\u8469\u95ee\u9898<\/strong>\u3002\u4e3a\u4e86\u907f\u514dChatGPT\u67d3\u4e0a\u6076\u4e60\uff0c ChatGPT \u901a\u8fc7<strong>\u7b97\u6cd5\u5c4f\u853d<\/strong>\uff0c\u51cf\u5c11\u6709\u5bb3\u548c\u6b3a\u9a97\u6027\u7684\u8bad\u7ec3\u8f93\u5165\uff0c\u67e5\u8be2\u901a\u8fc7\u9002\u5ea6 API \u8fdb\u884c\u8fc7\u6ee4\uff0c\u5e76\u9a73\u56de\u6f5c\u5728\u7684\u79cd\u65cf\u4e3b\u4e49\u6216\u6027\u522b\u6b67\u89c6\u63d0\u793a\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">2, ChatGPT\/GPT\u7684\u539f\u7406<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">2.1 NLP<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">NLP\/NLU\u9886\u57df<strong>\u5df2\u77e5\u5c40\u9650<\/strong>\u5305\u62ec\u5bf9\u91cd\u590d\u6587\u672c\u3001\u5bf9\u9ad8\u5ea6\u4e13\u4e1a\u7684\u4e3b\u9898\u7684\u8bef\u89e3\uff0c\u4ee5\u53ca\u5bf9<strong>\u4e0a\u4e0b\u6587\u77ed\u8bed\u7684\u8bef\u89e3<\/strong>\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5bf9\u4e8e\u4eba\u7c7b\u6216AI\uff0c\u901a\u5e38\u9700\u63a5\u53d7\u591a\u5e74\u7684\u8bad\u7ec3\u624d\u80fd\u6b63\u5e38\u5bf9\u8bdd\u3002NLP\u7c7b\u6a21\u578b\u4e0d\u4ec5\u8981\u7406\u89e3\u5355\u8bcd\u7684\u542b\u4e49\uff0c\u8fd8\u8981\u7406\u89e3\u5982\u4f55\u9020\u53e5\u548c\u7ed9\u51fa\u4e0a\u4e0b\u6587\u6709\u610f\u4e49\u7684\u56de\u7b54\uff0c\u751a\u81f3\u4f7f\u7528\u5408\u9002\u7684\u4fda\u8bed\u548c\u4e13\u4e1a\u8bcd\u6c47\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"720\" height=\"384\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-15.png\"  class=\"wp-image-209\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe4\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe4\" \/><figcaption class=\"wp-element-caption\">NLP\u7684\u5e94\u7528\u9886\u57df<\/figcaption><\/figure>\n<\/div>\n\n\n<p class=\"wp-block-paragraph\">\u672c\u8d28\u4e0a\uff0c\u4f5c\u4e3aChatGPT\u57fa\u7840\u7684GPT-3\u6216GPT-3.5 \u662f\u4e00\u4e2a\u8d85\u5927\u7684\u7edf\u8ba1\u8bed\u8a00\u6a21\u578b\u6216\u987a\u5e8f\u6587\u672c\u9884\u6d4b\u6a21\u578b\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">2.2 GPT v.s.BERT<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u4e0eBERT\u6a21\u578b\u7c7b\u4f3c\uff0cChatGPT\u6216GPT-3.5\u90fd\u662f\u6839\u636e\u8f93\u5165\u8bed\u53e5\uff0c\u6839\u636e\u8bed\u8a00\/\u8bed\u6599\u6982\u7387\u6765\u81ea\u52a8\u751f\u6210\u56de\u7b54\u7684\u6bcf\u4e00\u4e2a\u5b57\uff08\u8bcd\u8bed\uff09\u3002\u4ece\u6570\u5b66\u6216\u4ece\u673a\u5668\u5b66\u4e60\u7684\u89d2\u5ea6\u6765\u770b\uff0c\u8bed\u8a00\u6a21\u578b\u662f\u5bf9\u8bcd\u8bed\u5e8f\u5217\u7684<strong>\u6982\u7387\u76f8\u5173\u6027\u5206\u5e03\u7684\u5efa\u6a21<\/strong>\uff0c\u5373\u5229\u7528\u5df2\u7ecf\u8bf4\u8fc7\u7684\u8bed\u53e5\uff08\u8bed\u53e5\u53ef\u4ee5\u89c6\u4e3a\u6570\u5b66\u4e2d\u7684\u5411\u91cf\uff09\u4f5c\u4e3a\u8f93\u5165\u6761\u4ef6\uff0c\u9884\u6d4b\u4e0b\u4e00\u4e2a\u65f6\u523b\u4e0d\u540c\u8bed\u53e5\u751a\u81f3\u8bed\u8a00\u96c6\u5408\u51fa\u73b0\u7684\u6982\u7387\u5206\u5e03\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ChatGPT \u4f7f\u7528\u6765\u81ea\u4eba\u7c7b\u53cd\u9988\u7684\u5f3a\u5316\u5b66\u4e60\u8fdb\u884c\u8bad\u7ec3\uff0c\u8fd9\u79cd\u65b9\u6cd5\u901a\u8fc7\u4eba\u7c7b\u5e72\u9884\u6765\u589e\u5f3a\u673a\u5668\u5b66\u4e60\u4ee5\u83b7\u5f97\u66f4\u597d\u7684\u6548\u679c\u3002\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u4eba\u7c7b\u8bad\u7ec3\u8005\u626e\u6f14\u7740\u7528\u6237\u548c\u4eba\u5de5\u667a\u80fd\u52a9\u624b\u7684\u89d2\u8272\uff0c\u5e76\u901a\u8fc7<strong>\u8fd1\u7aef\u7b56\u7565\u4f18\u5316\u7b97\u6cd5<\/strong>\u8fdb\u884c\u5fae\u8c03\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7531\u4e8eChatGPT\u66f4\u5f3a\u7684\u6027\u80fd\u548c\u6d77\u91cf\u53c2\u6570\uff0c\u5b83\u5305\u542b\u4e86\u66f4\u591a\u7684\u4e3b\u9898\u7684\u6570\u636e\uff0c\u80fd\u591f\u5904\u7406\u66f4\u591a\u5c0f\u4f17\u4e3b\u9898\u3002ChatGPT\u73b0\u5728\u53ef\u4ee5\u8fdb\u4e00\u6b65\u5904\u7406\u56de\u7b54\u95ee\u9898\u3001\u64b0\u5199\u6587\u7ae0\u3001\u6587\u672c\u6458\u8981\u3001\u8bed\u8a00\u7ffb\u8bd1\u548c\u751f\u6210\u8ba1\u7b97\u673a\u4ee3\u7801\u7b49\u4efb\u52a1\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"720\" height=\"328\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-1.png\"  class=\"wp-image-189\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe5\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe5\" \/><figcaption class=\"wp-element-caption\">BERT\u4e0eGPT\u7684\u6280\u672f\u67b6\u6784\uff08\u56fe\u4e2dEn\u4e3a\u8f93\u5165\u7684\u6bcf\u4e2a\u5b57\uff0cTn\u4e3a\u8f93\u51fa\u56de\u7b54\u7684\u6bcf\u4e2a\u5b57\uff09<\/figcaption><\/figure>\n<\/div><\/div>\n<\/div>\n\n\n\n<h2 class=\"wp-block-heading\">3\uff0cChatGPT\u7684\u6280\u672f\u67b6\u6784<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">3.1 GPT\u5bb6\u65cf\u7684\u6f14\u8fdb<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u8bf4\u5230ChatGPT\uff0c\u5c31\u4e0d\u5f97\u4e0d\u63d0\u5230<strong>GPT\u5bb6\u65cf<\/strong>\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ChatGPT\u4e4b\u524d\u6709\u51e0\u4e2a\u77e5\u540d\u7684\u5144\u5f1f\uff0c\u5305\u62ecGPT-1\u3001GPT-2\u548cGPT-3\u3002\u8fd9\u51e0\u4e2a\u5144\u5f1f\u4e00\u4e2a\u6bd4\u4e00\u4e2a\u4e2a\u5934\u5927\uff0cChatGPT\u4e0eGPT-3\u66f4\u4e3a\u76f8\u8fd1\u3002<\/p>\n\n\n\n<div class=\"wp-block-cover aligncenter is-light\" style=\"min-height:500px;aspect-ratio:unset;\"><span aria-hidden=\"true\" class=\"wp-block-cover__background has-background-dim-0 has-background-dim\"><\/span><img decoding=\"async\" width=\"720\" height=\"438\" class=\"wp-block-cover__image-background wp-image-190\"  src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-2.png\" data-object-fit=\"cover\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe6\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe6\" \/><div class=\"wp-block-cover__inner-container is-layout-flow wp-block-cover-is-layout-flow\">\n<p class=\"has-text-align-center has-large-font-size wp-block-paragraph\">ChatGPT\u4e0eGPT 1-3\u7684\u6280\u672f\u5bf9\u6bd4<\/p>\n<\/div><\/div>\n\n\n\n<p class=\"wp-block-paragraph\">GPT\u5bb6\u65cf\u4e0eBERT\u6a21\u578b\u90fd\u662f\u77e5\u540d\u7684NLP\u6a21\u578b\uff0c\u90fd\u57fa\u4e8eTransformer\u6280\u672f\u3002GPT-1\u53ea\u6709<strong>12\u4e2a<\/strong>Transformer\u5c42\uff0c\u800c\u5230\u4e86GPT-3\uff0c\u5219\u589e\u52a0\u5230<strong>96\u5c42<\/strong>\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3.2 \u4eba\u7c7b\u53cd\u9988\u5f3a\u5316\u5b66\u4e60<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">InstructGPT\/GPT3.5\uff08ChatGPT\u7684\u524d\u8eab\uff09\u4e0eGPT-3\u7684\u4e3b\u8981\u533a\u522b\u5728\u4e8e\uff0c\u65b0\u52a0\u5165\u4e86\u88ab\u79f0\u4e3a<strong>RLHF<\/strong>\uff08Reinforcement Learning from Human Feedback\uff0c\u4eba\u7c7b\u53cd\u9988\u5f3a\u5316\u5b66\u4e60\uff09\u3002\u8fd9\u4e00\u8bad\u7ec3\u8303\u5f0f\u589e\u5f3a\u4e86\u4eba\u7c7b\u5bf9\u6a21\u578b\u8f93\u51fa\u7ed3\u679c\u7684\u8c03\u8282\uff0c\u5e76\u4e14\u5bf9\u7ed3\u679c\u8fdb\u884c\u4e86\u66f4\u5177\u7406\u89e3\u6027\u7684\u6392\u5e8f\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728InstructGPT\u4e2d\uff0c\u4ee5\u4e0b\u662f\u201cgoodness of sentences\u201d\u7684\u8bc4\u4ef7\u6807\u51c6\u3002<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u771f\u5b9e\u6027\uff1a\u662f\u865a\u5047\u4fe1\u606f\u8fd8\u662f\u8bef\u5bfc\u6027\u4fe1\u606f\uff1f<\/li>\n\n\n\n<li>\u65e0\u5bb3\u6027\uff1a\u5b83\u662f\u5426\u5bf9\u4eba\u6216\u73af\u5883\u9020\u6210\u8eab\u4f53\u6216\u7cbe\u795e\u4e0a\u7684\u4f24\u5bb3\uff1f<\/li>\n\n\n\n<li>\u6709\u7528\u6027\uff1a\u5b83\u662f\u5426\u89e3\u51b3\u4e86\u7528\u6237\u7684\u4efb\u52a1\uff1f<\/li>\n<\/ol>\n\n\n\n<h3 class=\"wp-block-heading\">3.3 TAMER\u6846\u67b6<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u8fd9\u91cc\u4e0d\u5f97\u4e0d\u63d0\u5230<strong>TAMER<\/strong>\uff08Training an Agent Manually via Evaluative Reinforcement\uff0c\u8bc4\u4f30\u5f0f\u5f3a\u5316\u4eba\u5de5\u8bad\u7ec3\u4ee3\u7406\uff09\u8fd9\u4e2a\u6846\u67b6\u3002\u8be5\u6846\u67b6\u5c06\u4eba\u7c7b\u6807\u8bb0\u8005\u5f15\u5165\u5230Agents\u7684\u5b66\u4e60\u5faa\u73af\u4e2d\uff0c\u53ef\u4ee5\u901a\u8fc7\u4eba\u7c7b\u5411Agents\u63d0\u4f9b<strong>\u5956\u52b1\u53cd\u9988<\/strong>\uff08\u5373\u6307\u5bfcAgents\u8fdb\u884c\u8bad\u7ec3\uff09\uff0c\u4ece\u800c\u5feb\u901f\u8fbe\u5230\u8bad\u7ec3\u4efb\u52a1\u76ee\u6807\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"946\" height=\"449\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-3.png\"  class=\"wp-image-191\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe7\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe7\" \/><figcaption class=\"wp-element-caption\">TAMER\u6846\u67b6\u8bba\u6587<\/figcaption><\/figure>\n<\/div>\n\n\n<p class=\"wp-block-paragraph\">\u5f15\u5165\u4eba\u7c7b\u6807\u8bb0\u8005\u7684\u4e3b\u8981\u76ee\u7684\u662f<strong>\u52a0\u5feb\u8bad\u7ec3\u901f\u5ea6<\/strong>\u3002\u5c3d\u7ba1\u5f3a\u5316\u5b66\u4e60\u6280\u672f\u5728\u5f88\u591a\u9886\u57df\u6709\u7a81\u51fa\u8868\u73b0\uff0c\u4f46\u662f\u4ecd\u7136\u5b58\u5728\u7740\u8bb8\u591a\u4e0d\u8db3\uff0c\u4f8b\u5982<strong>\u8bad\u7ec3\u6536\u655b\u901f\u5ea6\u6162\uff0c\u8bad\u7ec3\u6210\u672c\u9ad8<\/strong>\u7b49\u7279\u70b9\u3002\u7279\u522b\u662f\u73b0\u5b9e\u4e16\u754c\u4e2d\uff0c\u8bb8\u591a\u4efb\u52a1\u7684\u63a2\u7d22\u6210\u672c\u6216\u6570\u636e\u83b7\u53d6\u6210\u672c\u5f88\u9ad8\u3002\u5982\u4f55\u52a0\u5feb\u8bad\u7ec3\u6548\u7387\uff0c\u662f\u5982\u4eca\u5f3a\u5316\u5b66\u4e60\u4efb\u52a1\u5f85\u89e3\u51b3\u7684\u91cd\u8981\u95ee\u9898\u4e4b\u4e00\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u800cTAMER\u5219\u53ef\u4ee5\u5c06\u4eba\u7c7b\u6807\u8bb0\u8005\u7684\u77e5\u8bc6\uff0c\u4ee5\u5956\u52b1\u4fe1\u53cd\u9988\u7684\u5f62\u5f0f\u8bad\u7ec3Agent\uff0c\u52a0\u5feb\u5176\u5feb\u901f\u6536\u655b\u3002TAMER\u4e0d\u9700\u8981\u6807\u8bb0\u8005\u5177\u6709<strong>\u4e13\u4e1a\u77e5\u8bc6<\/strong>\u6216\u7f16\u7a0b\u6280\u672f\uff0c\u8bed\u6599\u6210\u672c\u66f4\u4f4e\u3002\u901a\u8fc7TAMER+RL\uff08\u5f3a\u5316\u5b66\u4e60\uff09\uff0c\u501f\u52a9\u4eba\u7c7b\u6807\u8bb0\u8005\u7684\u53cd\u9988\uff0c\u80fd\u591f\u589e\u5f3a\u4ece<strong>\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b<\/strong>&nbsp;(MDP) \u5956\u52b1\u8fdb\u884c\u5f3a\u5316\u5b66\u4e60 (RL) \u7684\u8fc7\u7a0b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"1302\" height=\"1016\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-4.png\"  class=\"wp-image-192\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe8\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe8\" \/><figcaption class=\"wp-element-caption\">TAMER\u67b6\u6784\u5728\u5f3a\u5316\u5b66\u4e60\u4e2d\u7684\u5e94\u7528<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">\u5177\u4f53\u5b9e\u73b0\u4e0a\uff0c\u4eba\u7c7b\u6807\u8bb0\u8005\u626e\u6f14\u5bf9\u8bdd\u7684\u7528\u6237\u548c\u4eba\u5de5\u667a\u80fd\u52a9\u624b\uff0c\u63d0\u4f9b\u5bf9\u8bdd\u6837\u672c\uff0c\u8ba9\u6a21\u578b\u751f\u6210\u4e00\u4e9b\u56de\u590d\uff0c\u7136\u540e\u6807\u8bb0\u8005\u4f1a\u5bf9\u56de\u590d\u9009\u9879<strong>\u6253\u5206\u6392\u540d<\/strong>\uff0c\u5c06\u66f4\u597d\u7684\u7ed3\u679c\u53cd\u9988\u56de\u6a21\u578b\u4e2d\uff0cAgents\u540c\u65f6\u4ece\u4e24\u79cd\u53cd\u9988\u6a21\u5f0f\u4e2d\u5b66\u4e60\u2014\u2014\u4eba\u7c7b\u5f3a\u5316\u548c\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\u5956\u52b1\u4f5c\u4e3a\u4e00\u4e2a\u6574\u5408\u7684\u7cfb\u7edf\uff0c\u901a\u8fc7<strong>\u5956\u52b1\u7b56\u7565<\/strong>\u5bf9\u6a21\u578b\u8fdb\u884c\u5fae\u8c03\u5e76\u6301\u7eed\u8fed\u4ee3\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5728\u6b64\u57fa\u7840\u4e0a\uff0cChatGPT \u53ef\u4ee5\u6bd4 GPT-3 \u66f4\u597d\u7684\u7406\u89e3\u548c\u5b8c\u6210\u4eba\u7c7b\u8bed\u8a00\u6216\u6307\u4ee4\uff0c\u6a21\u4eff\u4eba\u7c7b\uff0c\u63d0\u4f9b\u8fde\u8d2f\u7684\u6709\u903b\u8f91\u7684\u6587\u672c\u4fe1\u606f\u7684\u80fd\u529b\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3.4 ChatGPT\u7684\u8bad\u7ec3<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">ChatGPT\u7684\u8bad\u7ec3\u8fc7\u7a0b\u5206\u4e3a\u4ee5\u4e0b\u4e09\u4e2a\u9636\u6bb5\uff1a<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u7b2c\u4e00\u9636\u6bb5<\/strong>\uff1a\u8bad\u7ec3<strong>\u76d1\u7763\u7b56\u7565<\/strong>\u6a21\u578b<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">GPT 3.5\u672c\u8eab\u5f88\u96be\u7406\u89e3\u4eba\u7c7b\u4e0d\u540c\u7c7b\u578b\u6307\u4ee4\u4e2d\u8574\u542b\u7684\u4e0d\u540c\u610f\u56fe\uff0c\u4e5f\u5f88\u96be\u5224\u65ad\u751f\u6210\u5185\u5bb9\u662f\u5426\u662f\u9ad8\u8d28\u91cf\u7684\u7ed3\u679c\u3002\u4e3a\u4e86\u8ba9GPT 3.5\u521d\u6b65\u5177\u5907\u7406\u89e3\u6307\u4ee4\u7684\u610f\u56fe\uff0c\u9996\u5148\u4f1a\u5728\u6570\u636e\u96c6\u4e2d\u968f\u673a\u62bd\u53d6\u95ee\u9898\uff0c\u7531\u4eba\u7c7b\u6807\u6ce8\u4eba\u5458\uff0c\u7ed9\u51fa\u9ad8\u8d28\u91cf\u7b54\u6848\uff0c\u7136\u540e\u7528\u8fd9\u4e9b\u4eba\u5de5\u6807\u6ce8\u597d\u7684\u6570\u636e\u6765<strong>\u5fae\u8c03 GPT-3.5\u6a21\u578b<\/strong>\uff08\u83b7\u5f97SFT\u6a21\u578b, Supervised Fine-Tuning\uff09\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6b64\u65f6\u7684SFT\u6a21\u578b\u5728\u9075\u5faa\u6307\u4ee4\/\u5bf9\u8bdd\u65b9\u9762\u5df2\u7ecf\u4f18\u4e8e GPT-3\uff0c\u4f46\u4e0d\u4e00\u5b9a\u7b26\u5408\u4eba\u7c7b\u504f\u597d\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"720\" height=\"410\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-5.png\"  class=\"wp-image-193\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe9\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe9\" \/><figcaption class=\"wp-element-caption\">ChatGPT\u6a21\u578b\u7684\u8bad\u7ec3\u8fc7\u7a0b<\/figcaption><\/figure>\n<\/div>\n\n\n<p class=\"wp-block-paragraph\"><strong>\u7b2c\u4e8c\u9636\u6bb5<\/strong>\uff1a\u8bad\u7ec3<strong>\u5956\u52b1\u6a21\u578b<\/strong>\uff08Reward Mode\uff0cRM\uff09<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8fd9\u4e2a\u9636\u6bb5\u7684\u4e3b\u8981\u662f\u901a\u8fc7\u4eba\u5de5\u6807\u6ce8\u8bad\u7ec3\u6570\u636e\uff08\u7ea633K\u4e2a\u6570\u636e\uff09\uff0c\u6765\u8bad\u7ec3\u56de\u62a5\u6a21\u578b\u3002\u5728\u6570\u636e\u96c6\u4e2d\u968f\u673a\u62bd\u53d6\u95ee\u9898\uff0c\u4f7f\u7528\u7b2c\u4e00\u9636\u6bb5\u751f\u6210\u7684\u6a21\u578b\uff0c\u5bf9\u4e8e\u6bcf\u4e2a\u95ee\u9898\uff0c\u751f\u6210\u591a\u4e2a\u4e0d\u540c\u7684\u56de\u7b54\u3002\u4eba\u7c7b\u6807\u6ce8\u8005\u5bf9\u8fd9\u4e9b\u7ed3\u679c\u7efc\u5408\u8003\u8651\u7ed9\u51fa<strong>\u6392\u540d\u987a\u5e8f<\/strong>\u3002\u8fd9\u4e00\u8fc7\u7a0b\u7c7b\u4f3c\u4e8e\u6559\u7ec3\u6216\u8001\u5e08\u8f85\u5bfc\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u63a5\u4e0b\u6765\uff0c\u4f7f\u7528\u8fd9\u4e2a\u6392\u5e8f\u7ed3\u679c\u6570\u636e\u6765\u8bad\u7ec3\u5956\u52b1\u6a21\u578b\u3002\u5bf9\u591a\u4e2a\u6392\u5e8f\u7ed3\u679c\uff0c<strong>\u4e24\u4e24\u7ec4\u5408<\/strong>\uff0c\u5f62\u6210\u591a\u4e2a\u8bad\u7ec3\u6570\u636e\u5bf9\u3002RM\u6a21\u578b\u63a5\u53d7\u4e00\u4e2a\u8f93\u5165\uff0c\u7ed9\u51fa\u8bc4\u4ef7\u56de\u7b54\u8d28\u91cf\u7684\u5206\u6570\u3002\u8fd9\u6837\uff0c\u5bf9\u4e8e\u4e00\u5bf9\u8bad\u7ec3\u6570\u636e\uff0c\u8c03\u8282\u53c2\u6570\u4f7f\u5f97\u9ad8\u8d28\u91cf\u56de\u7b54\u7684\u6253\u5206\u6bd4\u4f4e\u8d28\u91cf\u7684\u6253\u5206\u8981\u9ad8\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>\u7b2c\u4e09\u9636\u6bb5<\/strong>\uff1a\u91c7\u7528<strong>PPO<\/strong>\uff08Proximal Policy Optimization\uff0c\u8fd1\u7aef\u7b56\u7565\u4f18\u5316\uff09<strong>\u5f3a\u5316\u5b66\u4e60\u6765\u4f18\u5316\u7b56\u7565<\/strong>\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">PPO\u7684\u6838\u5fc3\u601d\u8def\u5728\u4e8e\u5c06Policy Gradient\u4e2dOn-policy\u7684\u8bad\u7ec3\u8fc7\u7a0b\u8f6c\u5316\u4e3aOff-policy\uff0c\u5373\u5c06\u5728\u7ebf\u5b66\u4e60\u8f6c\u5316\u4e3a\u79bb\u7ebf\u5b66\u4e60\uff0c\u8fd9\u4e2a\u8f6c\u5316\u8fc7\u7a0b\u88ab\u79f0\u4e4b\u4e3a<strong>Importance Sampling<\/strong>\u3002\u8fd9\u4e00\u9636\u6bb5\u5229\u7528\u7b2c\u4e8c\u9636\u6bb5\u8bad\u7ec3\u597d\u7684\u5956\u52b1\u6a21\u578b\uff0c\u9760\u5956\u52b1\u6253\u5206\u6765\u66f4\u65b0\u9884\u8bad\u7ec3\u6a21\u578b\u53c2\u6570\u3002\u5728\u6570\u636e\u96c6\u4e2d\u968f\u673a\u62bd\u53d6\u95ee\u9898\uff0c\u4f7f\u7528PPO\u6a21\u578b\u751f\u6210\u56de\u7b54\uff0c\u5e76\u7528\u4e0a\u4e00\u9636\u6bb5\u8bad\u7ec3\u597d\u7684RM\u6a21\u578b\u7ed9\u51fa\u8d28\u91cf\u5206\u6570\u3002\u628a\u56de\u62a5\u5206\u6570\u4f9d\u6b21\u4f20\u9012\uff0c\u7531\u6b64\u4ea7\u751f\u7b56\u7565\u68af\u5ea6\uff0c\u901a\u8fc7\u5f3a\u5316\u5b66\u4e60\u7684\u65b9\u5f0f\u4ee5\u66f4\u65b0PPO\u6a21\u578b\u53c2\u6570\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5982\u679c\u6211\u4eec\u4e0d\u65ad\u91cd\u590d\u7b2c\u4e8c\u548c\u7b2c\u4e09\u9636\u6bb5\uff0c\u901a\u8fc7<strong>\u8fed\u4ee3<\/strong>\uff0c\u4f1a\u8bad\u7ec3\u51fa\u66f4\u9ad8\u8d28\u91cf\u7684ChatGPT\u6a21\u578b\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">4\uff0cChatGPT\u7684\u5c40\u9650<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">\u53ea\u8981\u7528\u6237\u8f93\u5165\u95ee\u9898\uff0cChatGPT \u5c31\u80fd\u7ed9\u4e88\u56de\u7b54\uff0c\u662f\u5426\u610f\u5473\u7740\u6211\u4eec\u4e0d\u7528\u518d\u62ff\u5173\u952e\u8bcd\u53bb\u5582 Google\u6216\u767e\u5ea6\uff0c\u5c31\u80fd\u7acb\u5373\u83b7\u5f97\u60f3\u8981\u7684\u7b54\u6848\u5462\uff1f<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5c3d\u7ba1ChatGPT\u8868\u73b0\u51fa\u51fa\u8272\u7684\u4e0a\u4e0b\u6587\u5bf9\u8bdd\u80fd\u529b\u751a\u81f3\u7f16\u7a0b\u80fd\u529b\uff0c\u5b8c\u6210\u4e86\u5927\u4f17\u5bf9\u4eba\u673a\u5bf9\u8bdd\u673a\u5668\u4eba\uff08ChatBot\uff09\u4ece\u201c\u4eba\u5de5\u667a\u969c\u201d\u5230\u201c\u6709\u8da3\u201d\u7684\u5370\u8c61\u6539\u89c2\uff0c\u6211\u4eec\u4e5f\u8981\u770b\u5230\uff0cChatGPT\u6280\u672f\u4ecd\u7136\u6709\u4e00\u4e9b\u5c40\u9650\u6027\uff0c\u8fd8\u5728\u4e0d\u65ad\u7684\u8fdb\u6b65\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">1\uff09ChatGPT\u5728\u5176\u672a\u7ecf<strong>\u5927\u91cf\u8bed\u6599<\/strong>\u8bad\u7ec3\u7684\u9886\u57df\u7f3a\u4e4f\u201c<strong>\u4eba\u7c7b\u5e38\u8bc6<\/strong>\u201d\u548c\u5f15\u7533\u80fd\u529b\uff0c\u751a\u81f3\u4f1a\u4e00\u672c\u6b63\u7ecf\u7684\u201c\u80e1\u8bf4\u516b\u9053\u201d\u3002ChatGPT\u5728\u5f88\u591a\u9886\u57df\u53ef\u4ee5\u201c<strong>\u521b\u9020\u7b54\u6848<\/strong>\u201d\uff0c\u4f46\u5f53\u7528\u6237\u5bfb\u6c42\u6b63\u786e\u7b54\u6848\u65f6\uff0cChatGPT\u4e5f\u6709\u53ef\u80fd\u7ed9\u51fa\u6709\u8bef\u5bfc\u7684\u56de\u7b54\u3002\u4f8b\u5982\u8ba9ChatGPT\u505a\u4e00\u9053\u5c0f\u5b66\u5e94\u7528\u9898\uff0c\u5c3d\u7ba1\u5b83\u53ef\u4ee5\u5199\u51fa\u4e00\u957f\u4e32\u8ba1\u7b97\u8fc7\u7a0b\uff0c\u4f46\u6700\u540e\u7b54\u6848\u5374\u662f\u9519\u8bef\u7684\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"640\" height=\"367\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-6.png\"  class=\"wp-image-194\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe10\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe10\" \/><figcaption class=\"wp-element-caption\">ChatGPT\u7ed9\u51fa\u7684\u9519\u8bef\u7684\u6570\u5b66\u9898\u7b54\u6848<\/figcaption><\/figure>\n<\/div>\n\n\n<p class=\"wp-block-paragraph\">2\uff09ChatGPT\u65e0\u6cd5\u5904\u7406\u590d\u6742\u5197\u957f\u6216\u8005<strong>\u7279\u522b\u4e13\u4e1a\u7684\u8bed\u8a00\u7ed3\u6784<\/strong>\u3002\u5bf9\u4e8e\u6765\u81ea\u91d1\u878d\u3001\u81ea\u7136\u79d1\u5b66\u6216\u533b\u5b66\u7b49\u975e\u5e38\u4e13\u4e1a\u9886\u57df\u7684\u95ee\u9898\uff0c\u5982\u679c\u6ca1\u6709\u8fdb\u884c\u8db3\u591f\u7684\u8bed\u6599\u201c\u5582\u98df\u201d\uff0cChatGPT\u53ef\u80fd\u65e0\u6cd5\u751f\u6210\u9002\u5f53\u7684\u56de\u7b54\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">3\uff09ChatGPT\u9700\u8981\u975e\u5e38<strong>\u5927\u91cf\u7684\u7b97\u529b\uff08\u82af\u7247\uff09<\/strong>\u6765\u652f\u6301\u5176\u8bad\u7ec3\u548c\u90e8\u7f72\u3002\u629b\u5f00\u9700\u8981\u5927\u91cf\u8bed\u6599\u6570\u636e\u8bad\u7ec3\u6a21\u578b\u4e0d\u8bf4\uff0c\u5728\u76ee\u524d\uff0cChatGPT\u5728\u5e94\u7528\u65f6\u4ecd\u7136\u9700\u8981\u5927\u7b97\u529b\u7684\u670d\u52a1\u5668\u652f\u6301\uff0c\u800c\u8fd9\u4e9b\u670d\u52a1\u5668\u7684\u6210\u672c\u662f\u666e\u901a\u7528\u6237\u65e0\u6cd5\u627f\u53d7\u7684\uff0c\u5373\u4fbf\u6570\u5341\u4ebf\u4e2a\u53c2\u6570\u7684\u6a21\u578b\u4e5f\u9700\u8981\u60ca\u4eba\u6570\u91cf\u7684<strong>\u8ba1\u7b97\u8d44\u6e90<\/strong>\u624d\u80fd\u8fd0\u884c\u548c\u8bad\u7ec3\u3002\u5982\u679c\u9762\u5411\u771f\u5b9e\u641c\u7d22\u5f15\u64ce\u7684\u6570\u4ee5\u4ebf\u8bb0\u7684\u7528\u6237\u8bf7\u6c42\uff0c\u5982\u91c7\u53d6\u76ee\u524d\u901a\u884c\u7684\u514d\u8d39\u7b56\u7565\uff0c\u4efb\u4f55\u4f01\u4e1a\u90fd\u96be\u4ee5\u627f\u53d7\u8fd9\u4e00\u6210\u672c\u3002\u56e0\u6b64\u5bf9\u4e8e\u666e\u901a\u5927\u4f17\u6765\u8bf4\uff0c\u8fd8\u9700\u7b49\u5f85\u66f4\u8f7b\u91cf\u578b\u7684\u6a21\u578b\u6216<strong>\u66f4\u9ad8\u6027\u4ef7\u6bd4\u7684\u7b97\u529b\u5e73\u53f0<\/strong>\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">4\uff09ChatGPT\u8fd8\u6ca1\u6cd5\u5728\u7ebf\u7684\u628a<strong>\u65b0\u77e5\u8bc6<\/strong>\u7eb3\u5165\u5176\u4e2d\uff0c\u800c\u51fa\u73b0\u4e00\u4e9b\u65b0\u77e5\u8bc6\u5c31\u53bb\u91cd\u65b0\u9884\u8bad\u7ec3GPT\u6a21\u578b\u4e5f\u662f\u4e0d\u73b0\u5b9e\u7684\uff0c\u65e0\u8bba\u662f\u8bad\u7ec3\u65f6\u95f4\u6216\u8bad\u7ec3\u6210\u672c\uff0c\u90fd\u662f\u666e\u901a\u8bad\u7ec3\u8005\u96be\u4ee5\u63a5\u53d7\u7684\u3002\u5982\u679c\u5bf9\u4e8e\u65b0\u77e5\u8bc6\u91c7\u53d6\u5728\u7ebf\u8bad\u7ec3\u7684\u6a21\u5f0f\uff0c\u770b\u4e0a\u53bb\u53ef\u884c\u4e14\u8bed\u6599\u6210\u672c\u76f8\u5bf9\u8f83\u4f4e\uff0c\u4f46\u662f\u5f88\u5bb9\u6613\u7531\u4e8e\u65b0\u6570\u636e\u7684\u5f15\u5165\u800c\u5bfc\u81f4\u5bf9\u539f\u6709\u77e5\u8bc6\u7684\u707e\u96be\u6027\u9057\u5fd8\u7684\u95ee\u9898\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">5\uff09ChatGPT\u4ecd\u7136\u662f<strong>\u9ed1\u76d2\u6a21\u578b<\/strong>\u3002\u76ee\u524d\u8fd8\u672a\u80fd\u5bf9ChatGPT\u7684\u5185\u5728\u7b97\u6cd5\u903b\u8f91\u8fdb\u884c\u5206\u89e3\uff0c\u56e0\u6b64\u5e76\u4e0d\u80fd\u4fdd\u8bc1ChatGPT\u4e0d\u4f1a\u4ea7\u751f\u653b\u51fb\u751a\u81f3\u4f24\u5bb3\u7528\u6237\u7684\u8868\u8ff0\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u5f53\u7136\uff0c<strong>\u7455\u4e0d\u63a9\u745c<\/strong>\uff0c\u6709\u5de5\u7a0b\u5e08\u8d34\u51fa\u4e86\u8981\u6c42ChatGPT\u5199verilog\u4ee3\u7801\uff08\u82af\u7247\u8bbe\u8ba1\u4ee3\u7801\uff09\u7684\u5bf9\u8bdd\u3002\u53ef\u4ee5\u770b\u51faChatGPT\u6c34\u5e73\u5df2\u7ecf\u8d85\u51fa\u4e00\u4e9bverilog\u521d\u5b66\u8005\u4e86\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">5\uff0cChatGPT\u7684\u672a\u6765\u6539\u8fdb\u65b9\u5411<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">5.1 \u51cf\u5c11\u4eba\u7c7b\u53cd\u9988\u7684RLAIF<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">2020\u5e74\u5e95\uff0cOpenAI\u524d\u7814\u7a76\u526f\u603b\u88c1Dario Amodei\u5e26\u774010\u540d\u5458\u5de5\u521b\u529e\u4e86\u4e00\u4e2a\u4eba\u5de5\u667a\u80fd\u516c\u53f8Anthropic\u3002Anthropic \u7684\u521b\u59cb\u56e2\u961f\u6210\u5458\uff0c\u5927\u591a\u4e3a OpenAI \u7684\u65e9\u671f\u53ca\u6838\u5fc3\u5458\u5de5\uff0c\u53c2\u4e0e\u8fc7OpenAI\u7684GPT-3\u3001\u591a\u6a21\u6001\u795e\u7ecf\u5143\u3001\u4eba\u7c7b\u504f\u597d\u7684\u5f3a\u5316\u5b66\u4e60\u7b49\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">2022\u5e7412\u6708\uff0cAnthropic\u518d\u6b21\u53d1\u8868\u8bba\u6587\u300aConstitutional AI: Harmlessness from AI Feedback\u300b\u4ecb\u7ecd\u4eba\u5de5\u667a\u80fd\u6a21\u578bClaude\u3002\uff08<a href=\"https:\/\/arxiv.org\/pdf\/2212.08073.pdf\">arxiv.org\/pdf\/2212.0807<\/a>\uff09<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img decoding=\"async\" width=\"812\" height=\"347\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-7.png\"  class=\"wp-image-195\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe11\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe11\" \/><figcaption class=\"wp-element-caption\">CAI\u6a21\u578b\u8bad\u7ec3\u8fc7\u7a0b<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Claude \u548c ChatGPT \u90fd\u4f9d\u8d56\u4e8e\u5f3a\u5316\u5b66\u4e60(RL)\u6765\u8bad\u7ec3\u504f\u597d\uff08preference\uff09\u6a21\u578b\u3002CAI\uff08Constitutional AI\uff09\u4e5f\u662f\u5efa\u7acb\u5728RLHF\u7684\u57fa\u7840\u4e4b\u4e0a\uff0c\u4e0d\u540c\u4e4b\u5904\u5728\u4e8e\uff0cCAI\u7684\u6392\u5e8f\u8fc7\u7a0b\u4f7f\u7528\u6a21\u578b\uff08\u800c\u975e\u4eba\u7c7b\uff09\u5bf9\u6240\u6709\u751f\u6210\u7684\u8f93\u51fa\u7ed3\u679c\u63d0\u4f9b\u4e00\u4e2a\u521d\u59cb\u6392\u5e8f\u7ed3\u679c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">CAI\u7528\u4eba\u5de5\u667a\u80fd\u53cd\u9988\u6765<strong>\u4ee3\u66ff\u4eba\u7c7b\u5bf9\u8868\u8fbe\u65e0\u5bb3\u6027\u7684\u504f\u597d<\/strong>\uff0c\u5373RLAIF\uff0c\u4eba\u5de5\u667a\u80fd\u6839\u636e\u4e00\u5957constitution\u539f\u5219\u6765\u8bc4\u4ef7\u56de\u590d\u5185\u5bb9\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"656\" height=\"390\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-8.png\"  class=\"wp-image-196\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe12\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe12\" \/><\/figure>\n<\/div>\n\n\n<h3 class=\"wp-block-heading\">5.2 \u8865\u8db3\u6570\u7406\u77ed\u677f<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">ChatGPT\u867d\u7136\u5bf9\u8bdd\u80fd\u529b\u5f3a\uff0c\u4f46\u662f\u5728\u6570\u7406\u8ba1\u7b97\u5bf9\u8bdd\u4e2d\u5bb9\u6613\u51fa\u73b0\u4e00\u672c\u6b63\u7ecf\u80e1\u8bf4\u516b\u9053\u7684\u60c5\u51b5\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u8ba1\u7b97\u673a\u5b66\u5bb6Stephen Wolfram \u4e3a\u8fd9\u4e00\u95ee\u9898\u63d0\u51fa\u4e86\u89e3\u51b3\u65b9\u6848\u3002Stephen Wolfram \u521b\u9020\u4e86\u7684 Wolfram \u8bed\u8a00\u548c\u8ba1\u7b97\u77e5\u8bc6\u641c\u7d22\u5f15\u64ce Wolfram | Alpha\uff0c\u5176\u540e\u53f0\u901a\u8fc7Mathematica\u5b9e\u73b0\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"720\" height=\"480\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-9.png\"  class=\"wp-image-197\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe13\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe13\" \/><figcaption class=\"wp-element-caption\">ChatGPT\u4e0eWolfram | Alpha\u7ed3\u5408\u5904\u7406\u68b3\u7406\u95ee\u9898<\/figcaption><\/figure>\n<\/div>\n\n\n<p class=\"wp-block-paragraph\">\u5728\u8fd9\u4e00\u7ed3\u5408\u4f53\u7cfb\u4e2d\uff0cChatGPT \u53ef\u4ee5\u50cf\u4eba\u7c7b\u4f7f\u7528 Wolfram|Alpha \u4e00\u6837\uff0c\u4e0e Wolfram|Alpha \u201c\u5bf9\u8bdd\u201d\uff0cWolfram|Alpha \u5219\u4f1a\u7528\u5176<strong>\u7b26\u53f7\u7ffb\u8bd1\u80fd\u529b<\/strong>\u5c06\u4ece ChatGPT \u83b7\u5f97\u7684\u81ea\u7136\u8bed\u8a00\u8868\u8fbe\u201c\u7ffb\u8bd1\u201d\u4e3a\u5bf9\u5e94\u7684\u7b26\u53f7\u5316\u8ba1\u7b97\u8bed\u8a00\u3002\u5728\u8fc7\u53bb\uff0c\u5b66\u672f\u754c\u5728 ChatGPT \u4f7f\u7528\u7684\u8fd9\u7c7b \u201c\u7edf\u8ba1\u65b9\u6cd5\u201d \u548c Wolfram|Alpha \u7684 \u201c\u7b26\u53f7\u65b9\u6cd5\u201d \u4e0a\u4e00\u76f4\u5b58\u5728\u8def\u7ebf\u5206\u6b67\u3002\u4f46\u5982\u4eca ChatGPT \u548c Wolfram|Alpha \u7684\u4e92\u8865\uff0c\u7ed9NLP\u9886\u57df\u63d0\u4f9b\u4e86\u66f4\u4e0a\u4e00\u5c42\u697c\u7684\u53ef\u80fd\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ChatGPT \u4e0d\u5fc5\u751f\u6210\u8fd9\u6837\u7684\u4ee3\u7801\uff0c\u53ea\u9700\u751f\u6210\u5e38\u89c4\u81ea\u7136\u8bed\u8a00\uff0c\u7136\u540e\u4f7f\u7528 Wolfram|Alpha \u7ffb\u8bd1\u6210\u7cbe\u786e\u7684 Wolfram Language\uff0c\u518d\u7531\u5e95\u5c42\u7684Mathematica\u8fdb\u884c\u8ba1\u7b97\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">5.3 ChatGPT\u7684\u5c0f\u578b\u5316<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u867d\u7136ChatGPT\u5f88\u5f3a\u5927\uff0c\u4f46\u5176\u6a21\u578b\u5927\u5c0f\u548c\u4f7f\u7528\u6210\u672c\u4e5f\u8ba9\u5f88\u591a\u4eba\u671b\u800c\u5374\u6b65\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6709\u4e09\u7c7b<strong>\u6a21\u578b\u538b\u7f29<\/strong>\uff08model compression\uff09\u53ef\u4ee5\u964d\u4f4e\u6a21\u578b\u7684\u5927\u5c0f\u548c\u6210\u672c\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7b2c\u4e00\u79cd\u65b9\u6cd5\u662f<strong>\u91cf\u5316<\/strong>\uff08quantization\uff09\uff0c\u5373\u964d\u4f4e\u5355\u4e2a\u6743\u91cd\u7684\u6570\u503c\u8868\u793a\u7684\u7cbe\u5ea6\u3002\u6bd4\u5982Tansformer\u4eceFP32\u964d\u5230INT8\u5bf9\u5176\u7cbe\u5ea6\u5f71\u54cd\u4e0d\u5927\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7b2c\u4e8c\u79cd\u6a21\u578b\u538b\u7f29\u65b9\u6cd5\u662f<strong>\u526a\u679d<\/strong>\uff08pruning\uff09\uff0c\u5373\u5220\u9664\u7f51\u7edc\u5143\u7d20\uff0c\u5305\u62ec\u4ece\u5355\u4e2a\u6743\u91cd\uff08\u975e\u7ed3\u6784\u5316\u526a\u679d\uff09\u5230\u66f4\u9ad8\u7c92\u5ea6\u7684\u7ec4\u4ef6\u5982\u6743\u91cd\u77e9\u9635\u7684\u901a\u9053\u3002\u8fd9\u79cd\u65b9\u6cd5\u5728\u89c6\u89c9\u548c\u8f83\u5c0f\u89c4\u6a21\u7684\u8bed\u8a00\u6a21\u578b\u4e2d\u6709\u6548\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u7b2c\u4e09\u79cd\u6a21\u578b\u538b\u7f29\u65b9\u6cd5\u662f<strong>\u7a00\u758f\u5316<\/strong>\u3002\u4f8b\u5982\u5965\u5730\u5229\u79d1\u5b66\u6280\u672f\u7814\u7a76\u6240 (ISTA)\u63d0\u51fa\u7684SparseGPT \uff08<a href=\"https:\/\/arxiv.org\/pdf\/2301.00774.pdf\">arxiv.org\/pdf\/2301.0077<\/a>\uff09\u53ef\u4ee5\u5c06 GPT \u7cfb\u5217\u6a21\u578b\u5355\u6b21\u526a\u679d\u5230 50% \u7684\u7a00\u758f\u6027\uff0c\u800c\u65e0\u9700\u4efb\u4f55\u91cd\u65b0\u8bad\u7ec3\u3002\u5bf9 GPT-175B \u6a21\u578b\uff0c\u53ea\u9700\u8981\u4f7f\u7528\u5355\u4e2a GPU \u5728\u51e0\u4e2a\u5c0f\u65f6\u5185\u5c31\u80fd\u5b9e\u73b0\u8fd9\u79cd\u526a\u679d\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"720\" height=\"309\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-10.png\"  class=\"wp-image-198\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe14\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe14\" \/><figcaption class=\"wp-element-caption\">SparseGPT \u538b\u7f29\u6d41\u7a0b<\/figcaption><\/figure>\n<\/div>\n\n\n<h2 class=\"wp-block-heading\">6, ChatGPT\u7684\u4ea7\u4e1a\u672a\u6765\u4e0e\u6295\u8d44\u673a\u4f1a<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">6.1 AIGC<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u8bf4\u5230ChaGPT\u4e0d\u5f97\u4e0d\u63d0AIGC\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>AIGC<\/strong>\u5373\u5229\u7528\u4eba\u5de5\u667a\u80fd\u6280\u672f\u6765\u751f\u6210\u5185\u5bb9\u3002\u4e0e\u6b64\u524dWeb1.0\u3001Web2.0\u65f6\u4ee3\u7684UGC\uff08\u7528\u6237\u751f\u4ea7\u5185\u5bb9\uff09\u548cPGC\uff08\u4e13\u4e1a\u751f\u4ea7\u5185\u5bb9\uff09\u76f8\u6bd4\uff0c\u4ee3\u8868\u4eba\u5de5\u667a\u80fd\u6784\u601d\u5185\u5bb9\u7684AIGC\uff0c\u662f\u65b0\u4e00\u8f6e\u5185\u5bb9<strong>\u751f\u4ea7\u65b9\u5f0f\u53d8\u9769<\/strong>\uff0c\u800c\u4e14AIGC\u5185\u5bb9\u5728Web3.0\u65f6\u4ee3\u4e5f\u5c06\u51fa\u73b0<strong>\u6307\u6570\u7ea7\u589e\u957f<\/strong>\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ChatGPT \u6a21\u578b\u7684\u51fa\u73b0\u5bf9\u4e8e\u6587\u5b57\/\u8bed\u97f3\u6a21\u6001\u7684 AIGC \u5e94\u7528\u5177\u6709\u91cd\u8981\u610f\u4e49\uff0c\u4f1a\u5bf9AI\u4ea7\u4e1a\u4e0a\u4e0b\u6e38<strong>\u4ea7\u751f\u91cd\u5927\u5f71\u54cd<\/strong>\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">6.2 \u53d7\u76ca\u573a\u666f<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">\u4ece\u4e0b\u6e38\u76f8\u5173\u53d7\u76ca\u5e94\u7528\u6765\u770b\uff0c\u5305\u62ec\u4f46\u4e0d\u9650\u4e8e<strong>\u65e0\u4ee3\u7801\u7f16\u7a0b<\/strong>\u3001\u5c0f\u8bf4\u751f\u6210\u3001\u5bf9\u8bdd\u7c7b\u641c\u7d22\u5f15\u64ce\u3001\u8bed\u97f3\u966a\u4f34\u3001\u8bed\u97f3\u5de5\u4f5c\u52a9\u624b\u3001\u5bf9\u8bdd\u865a\u62df\u4eba\u3001<strong>\u4eba\u5de5\u667a\u80fd\u5ba2\u670d<\/strong>\u3001\u673a\u5668\u7ffb\u8bd1\u3001<strong>\u82af\u7247\u8bbe\u8ba1<\/strong>\u7b49\u3002\u4ece\u4e0a\u6e38\u589e\u52a0\u9700\u6c42\u6765\u770b\uff0c\u5305\u62ec<strong>\u7b97\u529b\u82af\u7247<\/strong>\u3001\u6570\u636e\u6807\u6ce8\u3001\u81ea\u7136\u8bed\u8a00\u5904\u7406\uff08NLP)\u7b49\u3002<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"720\" height=\"448\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-11.png\"  class=\"wp-image-199\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe15\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe15\" \/><figcaption class=\"wp-element-caption\">\u5927\u6a21\u578b\u5448\u7206\u53d1\u6001\u52bf\uff08\u66f4\u591a\u7684\u53c2\u6570\/\u66f4\u5927\u7684\u7b97\u529b\u82af\u7247\u9700\u6c42\uff09<\/figcaption><\/figure>\n<\/div>\n\n\n<p class=\"wp-block-paragraph\">\u968f\u7740\u7b97\u6cd5\u6280\u672f\u548c\u7b97\u529b\u6280\u672f\u7684\u4e0d\u65ad\u8fdb\u6b65\uff0cChatGPT\u4e5f\u4f1a\u8fdb\u4e00\u6b65\u8d70\u5411\u66f4\u5148\u8fdb\u529f\u80fd\u66f4\u5f3a\u7684\u7248\u672c\uff0c\u5728\u8d8a\u6765\u8d8a\u591a\u7684\u9886\u57df\u8fdb\u884c\u5e94\u7528\uff0c\u4e3a\u4eba\u7c7b\u751f\u6210\u66f4\u591a\u66f4\u7f8e\u597d\u7684\u5bf9\u8bdd\u548c\u5185\u5bb9\u3002<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u6700\u540e\uff0c\u4f5c\u8005\u95ee<strong>\u5b58\u7b97\u4e00\u4f53<\/strong>\u6280\u672f\u5728ChatGPT\u9886\u57df\u7684\u5730\u4f4d\uff08\u4f5c\u8005\u672c\u4eba\u76ee\u524d\u5728\u91cd\u70b9\u63a8\u8fdb\u5b58\u7b97\u4e00\u4f53\u82af\u7247\u7684\u4ea7\u54c1\u843d\u5730\uff09\uff0cChatGPT\u60f3\u4e86\u60f3\uff0c\u5927\u80c6\u7684\u9884\u8a00\u5b58\u7b97\u4e00\u4f53\u6280\u672f\u5c06\u5728ChatGPT\u82af\u7247\u4e2d\u5360\u636e<strong>\u4e3b\u5bfc\u5730\u4f4d<\/strong>\u3002\uff08\u6df1\u5f97\u6211\u5fc3\uff09<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img decoding=\"async\" width=\"795\" height=\"196\" src=\"http:\/\/www.bosombot.com\/wp-content\/uploads\/2023\/02\/image-12.png\"  class=\"wp-image-200\" title=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe16\" alt=\"ChatGPT\u53d1\u5c55\u5386\u7a0b\u3001\u539f\u7406\u3001\u6280\u672f\u67b6\u6784\u8be6\u89e3\u548c\u4ea7\u4e1a\u672a\u6765\u63d2\u56fe16\" \/><\/figure>\n<\/div>\n\n\n<h2 class=\"wp-block-heading\">\u63a8\u8350\u6587\u7ae0<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"http:\/\/www.bosombot.com\/?p=117\">\u667a\u80fd\u751f\u6001\u7f51\u7edc(IEN)\uff1a\u77e5\u8bc6\u9a71\u52a8\u7684\u672a\u6765\u4ef7\u503c\u4e92\u8054\u7f51\u57fa\u7840\u8bbe\u65bd \u2013 \u77e5\u804a\u673a\u5668\u4eba\u793e\u533a (bosombot.com)<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"http:\/\/www.bosombot.com\/?p=155\">Web 3.0\u5e95\u5c42\u903b\u8f91\uff1a\u667a\u80fd\u751f\u6001\u7f51\u7edcIEN-\u5185\u5bb9\u4e2d\u5fc3\u5b89\u5168\u667a\u80fd\u94fe\u7f51\u878d\u5408\u67b6\u6784 \u2013 \u77e5\u804a\u673a\u5668\u4eba\u793e\u533a (bosombot.com)<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"http:\/\/www.bosombot.com\/?p=133\">Web3.0 \u5546\u4e1a\u6a21\u5f0f \u2014 \u667a\u80fd\u751f\u6001\u7f51\u7edcIEN \u2013 \u77e5\u804a\u673a\u5668\u4eba\u793e\u533a (bosombot.com)<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u6765\u6e90\uff1a\u9648\u5dcd\u8c08\u82af\uff0c\u672c\u6587\u5c06\u4ecb\u7ecdChatGPT\u7684\u7279\u70b9\u3001\u529f\u80fd\u3001\u6280&#8230; <\/p>\n","protected":false},"author":2,"featured_media":221,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[116],"tags":[],"class_list":["post-186","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-chanyejinzhan"],"_links":{"self":[{"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/posts\/186","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=186"}],"version-history":[{"count":12,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/posts\/186\/revisions"}],"predecessor-version":[{"id":245,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/posts\/186\/revisions\/245"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=\/wp\/v2\/media\/221"}],"wp:attachment":[{"href":"https:\/\/www.bosombot.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=186"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=186"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.bosombot.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=186"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}