{"id":50,"date":"2023-05-04T17:37:48","date_gmt":"2023-05-04T09:37:48","guid":{"rendered":"https:\/\/myhelpfulai.com\/?p=50"},"modified":"2023-05-04T18:41:25","modified_gmt":"2023-05-04T10:41:25","slug":"gpt-4%e5%a4%a7%e6%a8%a1%e5%9e%8b%e7%a1%ac%e6%a0%b8%e8%a7%a3%e8%af%bb","status":"publish","type":"post","link":"https:\/\/aitimes.link\/index.php\/2023\/05\/04\/gpt-4%e5%a4%a7%e6%a8%a1%e5%9e%8b%e7%a1%ac%e6%a0%b8%e8%a7%a3%e8%af%bb\/","title":{"rendered":"\u3010\u6458\u5f55\u3011GPT-4\u5927\u6a21\u578b\u786c\u6838\u89e3\u8bfb\uff01"},"content":{"rendered":"\n

GPT-4\u5927\u6a21\u578b\u786c\u6838\u89e3\u8bfb\uff01\u770b\u5b8c\u6210\u534a\u4e2a\u4e13\u5bb6
https:\/\/zhuanlan.zhihu.com\/p\/618761264<\/a><\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

1.1\u7406\u8bba\u57fa\u7840\u2014\u2014\u591a\u6a21\u6001\u6d8c\u73b0\u80fd\u529b<\/strong>\u8bb2\u5230\u5927\u8bed\u8a00\u6a21\u578b\u7684\u4f18\u52bf\uff0c\u4e00\u822c\u9996\u5148\u8981\u63d0\u5230\u8fd9\u7c7b\u6a21\u578b\u7684\u6d8c\u73b0\u80fd\u529b\u548c\u601d\u7ef4\u94fe\u3002\u8fd9\u4e24\u8005\u662f\u5927\u8bed\u8a00\u6a21\u578b\u4e0d\u65ad\u63a5\u8fd1\u4eba\u7c7b\u7684\u5173\u952e\u7279\u5f81<\/strong>\u3002\u6211\u4eec\u4e4b\u6240\u4ee5\u8ba4\u4e3aGPT-4\u4f1a\u662f\u5177\u6709\u91cc\u7a0b\u7891\u610f\u4e49\u7684\u4e00\u4ee3\uff0c\u6b63\u662f\u56e0\u4e3a\u591a\u6a21\u6001\u7684GPT-4\u4f1a\u4ece\u89c6\u89c9\u89d2\u5ea6\u548c\u89c6\u89c9-\u6587\u5b57\u8bed\u4e49\u878d\u5408\u65b9\u9762\u6d8c\u73b0\u51fa\u66f4\u591a\u7684\u80fd\u529b\u30022022-2023\u5e74\uff0c\u6211\u4eec\u53ef\u4ee5\u8ba4\u4e3aAI\u662f\u7b2c\u4e00\u6b21\u7741\u5f00\u53cc\u773c<\/strong>\u7406\u89e3\u8fd9\u4e2a\u4e16\u754c\u3002\u5728\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u4e2d\uff0c\u6d8c\u73b0\u80fd\u529b<\/strong>\uff08Emergent Abilities\uff09\u662f\u6307\u6a21\u578b\u5177\u6709\u4ece\u539f\u59cb\u8bad\u7ec3\u6570\u636e\u4e2d\u81ea\u52a8\u5b66\u4e60\u5e76\u53d1\u73b0\u65b0\u7684\u3001\u66f4\u9ad8\u5c42\u6b21\u7684\u7279\u5f81\u548c\u6a21\u5f0f\u7684\u80fd\u529b\u3002\u5c31\u4e2d\u6587\u91ca\u4e49\u800c\u8a00\uff0c\u6d8c\u73b0\u80fd\u529b\u4e5f\u6307\u5927\u8bed\u8a00\u6a21\u578b\u6d8c\u73b0\u51fa\u6765\u7684\u65b0\u80fd\u529b\u3002\u8fd9\u6709\u70b9\u7c7b\u4f3c\u4e8e\u53bb\u8d85\u5e02\u9047\u5230\u4e70\u4e8c\u8d60\u4e00\uff0c\u8d60\u54c1\u7684\u8d28\u91cf\u5c45\u7136\u8fd8\u51fa\u4e4e\u610f\u6599\u3002\u4e0e\u5927\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u76f8\u6bd4\uff0c\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b\uff08Multi-modal Large Language Model\uff0cMLLM\uff09\u53ef\u5b9e\u73b0\u66f4\u597d\u7684\u5e38\u8bc6\u63a8\u7406\u6027\u80fd\uff0c\u8de8\u6a21\u6001\u8fc1\u79fb\u66f4\u6709\u5229\u4e8e\u77e5\u8bc6\u83b7\u53d6\uff0c\u4ea7\u751f\u66f4\u591a\u65b0\u7684\u80fd\u529b\uff0c\u52a0\u901f\u4e86\u80fd\u529b\u7684\u6d8c\u73b0\u3002\u8fd9\u4e9b\u72ec\u7acb\u6a21\u6001\u6216\u8de8\u6a21\u6001\u65b0\u7279\u5f81\u3001\u80fd\u529b\u6216\u6a21\u5f0f\u901a\u5e38\u4e0d\u662f\u901a\u8fc7\u76ee\u7684\u660e\u786e<\/strong>\u7684\u7f16\u7a0b\u6216\u8bad\u7ec3\u83b7\u5f97\u7684\uff0c\u800c\u662f\u6a21\u578b\u5728\u5927\u91cf\u591a\u6a21\u6001\u6570\u636e\u4e2d\u81ea\u7136\u800c\u7136<\/strong>\u7684\u5b66\u4e60\u5230\u7684\u3002<\/p>\n\n\n\n

\u25b2\u7f29\u653e\u5b9a\u5f8b\uff08\u53c2\u6570\u589e\u52a0\u540e\u7cbe\u5ea6\u635f\u5931\u8fde\u7eed\u51cf\u5c11\uff09V.S. \u6d8c\u73b0\u80fd\u529b\uff081010-1011\u53c2\u6570\u540e\u65b0\u80fd\u529b\u7684\u6d8c\u73b0\uff09\uff08\u6765\u6e90\uff1aOpenAI\uff09<\/p>\n\n\n\n

\u25b2\u601d\u7ef4\u94fe\u63d0\u793a\u7684\u793a\u4f8b\uff08\u6765\u6e90\uff1aGoogle\uff09<\/p>\n\n\n\n

\u25b2\u591a\u6a21\u6001\u601d\u7ef4\u94fe\u6846\u67b6\uff08\u6765\u6e90\uff1a\u5fae\u8f6f\uff09<\/p>\n\n\n\n

\u25b2\u8bed\u8a00\u6a21\u578b\u76844\u79cd\u7814\u7a76\u8303\u5f0f\uff08\u6765\u6e90\uff1a\u5361\u5185\u57fa\u6885\u9686\u5927\u5b66\uff09<\/p>\n\n\n\n

\u5927\u6982\u57282017-2019\u5e74\u95f4\uff0c\u8bed\u8a00\u6a21\u578b\u7684\u7814\u7a76\u91cd\u5fc3\u9010\u6e10\u4ece\u4f20\u7edf\u7279\u5b9a\u9886\u57df\u7684\u6709\u76d1\u7763\u5b66\u4e60\u6a21\u5f0f\uff08\u57fa\u4e8e\u975e\u795e\u7ecf\u7f51\u7edc\u6216\u795e\u7ecf\u7f51\u7edc\uff09\u8f6c\u79fb\u5230\u9884\u8bad\u7ec3\u6a21\u578b\u4e0a\u3002\u5728\u90a3\u65f6\uff0c\u57fa\u4e8e\u9884\u8bad\u7ec3\u8bed\u8a00\u6a21\u578b\u7684\u7814\u7a76\u8303\u5f0f\u901a\u5e38\u662f\u201c\u9884\u8bad\u7ec3+\u7cbe\u8c03<\/strong>\u201d\uff08Pre-train+Fine-tune\uff09\uff0c\u5373\u5728\u7cbe\u8c03\u9636\u6bb5\uff0c\u6839\u636e\u4e0b\u6e38\u4efb\u52a1\u5bf9\u9884\u8bad\u7ec3\u6a21\u578b\u8fdb\u884c\u5fae\u8c03\uff0c\u4ee5\u83b7\u5f97\u66f4\u597d\u6548\u679c\u3002\u4f46\u662f\u7531\u4e8e\u6a21\u578b\u8d8a\u6765\u8d8a\u5927\uff0c\u4ee5\u53ca\u9884\u8bad\u7ec3\u9636\u6bb5\u548c\u4e0b\u6e38\u4efb\u52a1\u4e4b\u95f4\u7684\u5dee\u8ddd\u53ef\u80fd\u5f88\u5927\uff0c\u5bf9\u5404\u4e2a\u7ec6\u5206\u9886\u57dfFine-tune\u7684\u8ba1\u7b97\u8d44\u6e90\u8981\u6c42\u3001\u8bad\u7ec3\u6570\u636e\u9700\u6c42\u548c\u65f6\u95f4\u6210\u672c\u4e5f\u5728\u5feb\u901f\u4e0a\u6da8\u3002\u5927\u91cf\u7206\u53d1\u7684\u4e0b\u6e38\u4efb\u52a1\u4e5f\u4f7f\u5f97175B\u8fd9\u4e2a\u7ea7\u522b\u6a21\u578b\u9884\u8bad\u7ec3\u548c\u7cbe\u8c03\u53d8\u5f97\u5f02\u5e38\u590d\u6742\u3002\u5728\u8fd9\u79cd\u80cc\u666f\u4e0b\uff0c\u968f\u7740GPT-3\u7684\u53d1\u5e03\uff0c\u63d0\u793a\u5de5\u7a0b\u6210\u4e3a\u4e86\u9884\u8bad\u7ec3\u6a21\u578b\u7684\u65b0\u65b9\u5411\u3002\u5f62\u8c61\u7684\u8bf4\uff0c\u63d0\u793a\u6709\u70b9\u7c7b\u4f3c\u4e8e\u8001\u5e08\u5728\u5b66\u751f\u56de\u7b54\u95ee\u9898\u65f6\u6307\u70b9\u56de\u7b54\u65b9\u5411\u3002<\/p>\n\n\n\n

\u25b2\u5956\u52b1\u6a21\u578b\u7684\u8fc7\u62df\u5408\u5bfc\u81f4\u6a21\u578b\u6027\u80fd\u4e0b\u964d\uff08\u6765\u6e90\uff1aOpenAI\uff09<\/p>\n\n\n\n

\u25b2\u57fa\u4e8e\u89c4\u5219\u7684\u5956\u52b1\u6a21\u578b\uff08\u6765\u6e90\uff1a\u65e5\u672c\u56fd\u7acb\u4fe1\u606f\u5b66\u7814\u7a76\u6240\uff09<\/p>\n\n\n\n

\u25b2\u57fa\u4e8e\u89c4\u5219\u7684\u5956\u52b1\u6a21\u578b\u5728\u6837\u672c\u8f83\u5c11\u60c5\u51b5\u4e0b\u8868\u73b0\u51fa\u8f83\u597d\u6027\u80fd\uff08\u6765\u6e90\uff1aMeta AI\uff09<\/p>\n\n\n\n

\u25b2PPO\u7b97\u6cd5\u4e0e\u540c\u7c7b\u5176\u4ed6\u7b97\u6cd5\u7684\u6bd4\u8f83\uff08\u6765\u6e90\uff1aOpenAI\uff09<\/p>\n\n\n\n

\u25b2\u5927\u8bed\u8a00\u6a21\u578b\u53ef\u89c6\u4e3a\u77e5\u8bc6\/\u8bed\u8a00\u7684\u6709\u635f\u538b\u7f29<\/p>\n\n\n\n

\u25b2\u901a\u8fc7\u5e7b\u89c9\u5355\u8bcd\u68c0\u6d4b\u5668\u51cf\u5c11\u5e7b\u89c9\uff08\u6765\u6e90\uff1aMeta AI\uff09<\/p>\n\n\n\n

Bing\u53cd\u9988\u7684GPT-4\u6a21\u578b\u5927\u5c0f<\/p>\n\n\n\n

GPT-3\u662f\u76ee\u524d\u6700\u5927\u7684\u77e5\u540d\u8bed\u8a00\u6a21\u578b\u4e4b\u4e00\uff0c\u5305\u542b\u4e861750\u4ebf\uff08175B\uff09\u4e2a\u53c2\u6570\u3002\u5728GPT-3\u53d1\u5e03\u4e4b\u524d\uff0c\u6700\u5927\u7684\u8bed\u8a00\u6a21\u578b\u662f\u5fae\u8f6f\u7684Turing NLG\u6a21\u578b\uff0c\u5927\u5c0f\u4e3a17\u4ebf\uff081.7B\uff09\u4e2a\u53c2\u6570\u3002\u5728GPT-3\u53d1\u5e03\u540e\u4e0d\u4e45\uff0cOpenAI\u56e2\u961f\u5c31\u66fe\u8868\u793a\u4ed6\u4eec\u8ba1\u5212\u5728\u672a\u6765\u51e0\u5e74\u5185\u7814\u53d1\u66f4\u5927\u7684\u6a21\u578b\u3002\u800c\u968f\u7740\u6280\u672f\u548c\u7b97\u6cd5\u7684\u4e0d\u65ad\u53d1\u5c55\uff0cGPT-4\u6a21\u578b\u4f3c\u4e4e\u4e5f\u5e94\u671d\u7740\u66f4\u5927\u7684\u5c3a\u5bf8\u53d1\u5c55\u3002\u53e6\u5916\uff0cGPT-4\u7684\u4e0a\u4e0b\u6587\u7a97\u53e3\u5c3a\u5bf8\u4e5f\u8f83GPT-3.5\u548cGPT-3\u589e\u5927\u4e86\u4e0d\u5c11\u30022020\u5e74\u53d1\u5e03\u7684GPT-3\u6a21\u578b\u4e0a\u4e0b\u6587\u7a97\u53e3\u4e3a2049\u4e2a\u4ee4\u724c\u3002\u5728GPT-3.5\u4e2d\uff0c\u7a97\u53e3\u589e\u52a0\u52304096\u4e2a\u4ee4\u724c\uff08\u7ea63\u9875\u5355\u884c\u82f1\u6587\u6587\u672c\uff09\u3002GPT-4\u6709\u4e24\u79cd\u5c3a\u5bf8\u3002\u5176\u4e2d\u4e00\u4e2a\uff08GPT-4-8K\uff09\u7684\u4e0a\u4e0b\u6587\u7a97\u53e3\u5927\u5c0f\u4e3a8192\u4e2a\u4ee4\u724c\uff0c\u53e6\u4e00\u4e2a\uff08GPT-4-32K\uff09\u53ef\u4ee5\u5904\u7406\u591a\u8fbe32768\u4e2a\u4ee4\u724c\uff0c\u5927\u7ea650\u9875\u6587\u672c\u3002\u6709\u4f20\u8a00\u8bf4GPT-4\u6a21\u578b\u5927\u6982\u662fGPT-3\u7684100\u500d\u62161000\u500d\u3002\u4ece\u8bad\u7ec3\u7684\u89d2\u5ea6\u770b\uff0c\u8fd9\u4e48\u5927\u7684\u6a21\u578b\u81a8\u80c0\u53ef\u80fd\u4f1a\u6d88\u8017\u66f4\u591a\u7684\u8bad\u7ec3\u8d44\u6e90\u548c\u8bad\u7ec3\u5468\u671f\u7684\u8fc7\u5ea6\u5ef6\u957f\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

GPT-4\u5927\u6a21\u578b\u786c\u6838\u89e3\u8bfb\uff01\u770b\u5b8c\u6210\u534a\u4e2a\u4e13\u5bb6https:\/\/zhuanlan.zhihu.com\/p\/618761 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6,11],"tags":[],"class_list":["post-50","post","type-post","status-publish","format-standard","hentry","category-ai","category-11"],"_links":{"self":[{"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/posts\/50","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/comments?post=50"}],"version-history":[{"count":2,"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/posts\/50\/revisions"}],"predecessor-version":[{"id":57,"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/posts\/50\/revisions\/57"}],"wp:attachment":[{"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/media?parent=50"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/categories?post=50"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aitimes.link\/index.php\/wp-json\/wp\/v2\/tags?post=50"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}