{"id":55403,"date":"2025-02-04T21:04:21","date_gmt":"2025-02-04T13:04:21","guid":{"rendered":"https:\/\/swarma.org\/?p=55403"},"modified":"2025-02-04T21:04:21","modified_gmt":"2025-02-04T13:04:21","slug":"%e5%a4%9a%e6%a8%a1%e6%80%81%e5%a4%a7%e8%af%ad%e8%a8%80%e6%a8%a1%e5%9e%8b-vs-%e4%ba%ba%e7%b1%bb%ef%bc%9a%e8%a7%86%e8%a7%89%e8%ae%a4%e7%9f%a5%e8%83%bd%e5%8a%9b%e7%9a%84%e8%be%83%e9%87%8f","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=55403","title":{"rendered":"\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b vs \u4eba\u7c7b\uff1a\u89c6\u89c9\u8ba4\u77e5\u80fd\u529b\u7684\u8f83\u91cf"},"content":{"rendered":"
\n
\n
\n
\n
\n

<\/p>\n


<\/p>\n<\/section>\n

\n
\n
\n
\n
<\/strong><\/section>\n
\u5173\u952e\u8bcd\uff1a\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\uff0c\u76f4\u89c9\u7269\u7406\u5b66\uff0c\u56e0\u679c\u63a8\u7406\uff0c\u76f4\u89c9\u5fc3\u7406\u5b66\uff0c\u89c6\u89c9\u8ba4\u77e5<\/span><\/strong><\/span><\/strong><\/span><\/strong><\/section>\n

<\/span><\/strong><\/section>\n<\/section>\n
\n
<\/strong><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/p>\n


<\/p>\n

<\/span><\/section>\n


<\/h2>\n

<\/p>\n

\n

\u8bba\u6587\u9898\u76ee\uff1aVisual cognition in multimodal large language models<\/span><\/p>\n

\u8bba\u6587\u5730\u5740\uff1ahttps:\/\/www.nature.com\/articles\/s42256-024-00963-y<\/span><\/p>\n

\u671f\u520a\u540d\u79f0\uff1aNature Machine Intelligence<\/em><\/span><\/p>\n<\/blockquote>\n


<\/p>\n

\u968f\u7740\u5927\u8bed\u8a00\u6a21\u578b<\/span>\uff08Large Language Models, LLMs\uff09<\/span>\u7684\u5d1b\u8d77\uff0c\u7814\u7a76\u8005\u6bd4\u8f83\u4eba\u5de5\u667a\u80fd\u6a21\u578b\u548c\u4eba\u7c7b\u8ba4\u77e5\u80fd\u529b\u7684\u5174\u8da3\u518d\u6b21\u9ad8\u6da8\u3002\u4eba\u4eec\u666e\u904d\u8ba4\u4e3a\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u67b6\u6784\u65e0\u6cd5\u505a\u5230\u50cf\u4eba\u4e00\u6837\u601d\u8003\uff0c\u5728\u56e0\u679c\u63a8\u7406\u3001\u76f4\u89c9\u7269\u7406\u5b66\u548c\u76f4\u89c9\u5fc3\u7406\u5b66\u7b49\u9886\u57df\u7684\u8868\u73b0\u4e00\u76f4\u53d7\u5230\u8d28\u7591\uff0c\u4f46\u6700\u65b0\u7684\u591a\u6a21\u6001\u5927\u8bed\u8a00\u6a21\u578b<\/span>\uff08Multimodal Large Language Models, MLLMs\uff09<\/span>\u5c55\u793a\u4e86\u5728\u89c6\u89c9\u5904\u7406\u4e0a\u7684\u6f5c\u529b\uff0c\u672c\u6587\u901a\u8fc7\u4e00\u7cfb\u5217\u5b9e\u9a8c\u63a2\u8ba8MLLMs\u5728\u4e0a\u8ff0\u9886\u57df\u7684\u8868\u73b0\u3002<\/span><\/p>\n


<\/span><\/p>\n


<\/strong><\/span><\/p>\n

\n
\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

\u5b9e\u9a8c\u65b9\u6cd5\u4e0e\u53d1\u73b0<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n


<\/p>\n

\u7814\u7a76\u91c7\u7528\u4e00\u7cfb\u5217\u7ecf\u5178\u8ba4\u77e5\u79d1\u5b66\u4efb\u52a1\u6765\u8bc4\u4f30\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u7684\u8ba4\u77e5\u80fd\u529b\u3002<\/span><\/p>\n


<\/span><\/p>\n

<\/p>\n


<\/span><\/p>\n

\u5728\u76f4\u89c9\u7269\u7406\u611f<\/strong>\u65b9\u9762\uff0c\u4f7f\u7528\u4e86\u6765\u81eaLerer\u7b49\u4eba\u7684\u6728\u5757\u5854\u5b9e\u9a8c\uff0c\u8981\u6c42\u6a21\u578b\u5224\u65ad\u5854\u7684\u7a33\u5b9a\u6027\uff0c\u4ece\u800c\u8bc4\u5224\u6a21\u578b\u662f\u5426\u638c\u63e1\u4e86\u590d\u6742\u7684\u7269\u7406\u76f8\u4e92\u4f5c\u7528<\/strong>\u3002\u7ed3\u679c\u663e\u793a\uff0cGPT-4V\u548cClaude-3\u5728\u7a33\u5b9a\u6027\u5224\u65ad\u4e0a\u8868\u73b0\u7565\u9ad8\u4e8e\u968f\u673a\u6c34\u5e73\uff0c\u4f46\u4e0e\u4eba\u7c7b\u768465.6%\u51c6\u786e\u7387\u76f8\u6bd4\u4ecd\u6709\u5dee\u8ddd\u3002<\/span><\/p>\n


<\/span><\/p>\n

<\/p>\n


<\/span><\/p>\n

\u5728\u56e0\u679c\u63a8\u7406<\/strong>\u65b9\u9762\uff0c\u7814\u7a76\u91c7\u7528\u4e86Jenga\u79ef\u6728\u5b9e\u9a8c\u548cMichotte\u53d1\u5c04\u8303\u5f0f\uff0c\u8981\u6c42\u6a21\u578b\u5224\u65ad\u79ef\u6728\u7684\u56e0\u679c\u5173\u7cfb\uff0c\u4ee5\u8003\u9a8c\u5927\u6a21\u578b\u5bf9\u56e0\u679c\u5173\u7cfb\u7684\u7406\u89e3<\/strong>\u3002\u7ed3\u679c\u8868\u660e\uff0c\u5c3d\u7ba1\u6a21\u578b\u5728\u67d0\u4e9b\u4efb\u52a1\u4e2d\u8868\u73b0\u7565\u4f18\u4e8e\u968f\u673a\u6c34\u5e73\uff0c\u4f46\u548c\u4eba\u7c7b\u7684\u56e0\u679c\u5224\u65ad\u4e00\u81f4\u6027\u4e0a\u4ecd\u663e\u4e0d\u8db3\u3002<\/span><\/p>\n


<\/span><\/p>\n

<\/p>\n


<\/span><\/p>\n

\u5728\u76f4\u89c9\u5fc3\u7406<\/strong>\u65b9\u9762\uff0c\u7814\u7a76\u8bbe\u8ba1\u4e86\u5b87\u822a\u5458\u4efb\u52a1\u548c\u5e2e\u52a9\u6216\u963b\u788d\u4efb\u52a1\uff0c\u8981\u6c42\u6a21\u578b\u63a8\u65ad\u573a\u666f\u4e2d\u89d2\u8272\u7684\u610f\u56fe\u548c\u884c\u4e3a\u7ed3\u679c\uff0c\u8bc4\u4f30\u5176\u5bf9\u4ed6\u4eba\u504f\u597d\u7684\u76f4\u89c9\u7406\u89e3\u80fd\u529b<\/strong>\u3002\u7ed3\u679c\u663e\u793a\uff0c\u6240\u6709\u6a21\u578b\u5728\u4efb\u52a1\u590d\u6742\u6027\u9012\u589e\u65f6\u7684\u8868\u73b0\u5747\u4e0d\u53ca\u4eba\u7c7b\u3002\u6b64\u5916\uff0cGPT-4V\u5728\u4e2a\u522b\u4efb\u52a1\u4e2d\u8868\u73b0\u51fa\u4e0e\u4eba\u7c7b\u5224\u65ad\u7684\u5fae\u5f31\u76f8\u5173\u6027\uff0c\u4f46\u6574\u4f53\u4e0a\u4ecd\u672a\u8fbe\u5230\u4eba\u7c7b\u6c34\u5e73\u3002<\/span><\/p>\n


<\/span><\/p>\n

<\/p>\n


<\/p>\n


<\/span><\/p>\n

\n
\n
\n
\n
\n

<\/section>\n

<\/section>\n
\n
<\/svg><\/section>\n<\/section>\n<\/section>\n
\n

\u8ba8\u8bba\u4e0e\u7ed3\u8bba<\/strong><\/span><\/strong><\/strong><\/p>\n<\/section>\n

\n

<\/section>\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/h3>\n


<\/p>\n

\u5c3d\u7ba1\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u5728\u89c6\u89c9\u6570\u636e\u5904\u7406\u4e0a\u5c55\u73b0\u51fa\u4e00\u5b9a\u7684\u80fd\u529b\uff0c\u4f46\u5728\u548c\u4eba\u7c7b\u76f4\u89c9\u8ba4\u77e5\u5bf9\u6bd4\u4e0b\u4ecd\u6709\u663e\u8457\u4e0d\u8db3\u3002\u7814\u7a76\u6307\u51fa\uff0c\u6a21\u578b\u5728\u57fa\u7840\u89c6\u89c9\u5904\u7406\u4e0a\u7684\u5c40\u9650\u53ef\u80fd\u662f\u5bfc\u81f4\u5176\u8ba4\u77e5\u8868\u73b0\u4e0d\u4f73\u7684\u539f\u56e0\u4e4b\u4e00\uff0c\u4ee5\u53ca\u5b9e\u9a8c\u4e2d\u4f7f\u7528\u7684\u9759\u6001\u56fe\u50cf\u53ef\u80fd\u9650\u5236\u4e86\u4efb\u52a1\u7684\u590d\u6742\u6027\uff0c\u672a\u6765\u53ef\u4ee5\u8003\u8651\u4f7f\u7528\u66f4\u5177\u52a8\u6001\u6027\u548c\u590d\u6742\u6027\u7684\u523a\u6fc0\u6750\u6599\u3002\u7814\u7a76\u5f3a\u8c03\u4e86\u9700\u8981\u66f4\u4e3a\u5065\u5168\u7684\u56e0\u679c\u5173\u7cfb\u3001\u7269\u7406\u52a8\u6001\u548c\u793e\u4f1a\u8ba4\u77e5\u673a\u5236<\/strong>\uff0c\u4ee5\u63d0\u5347\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u7684\u8868\u73b0\u3002\u672a\u6765\u7684\u7814\u7a76\u53ef\u4ee5\u63a2\u7d22\u901a\u8fc7\u8c03\u6574\u6a21\u578b\u89c4\u6a21\u6216\u5fae\u8c03\u8bad\u7ec3\u8fc7\u7a0b\u6765\u6539\u5584\u6a21\u578b\u7684\u8ba4\u77e5\u80fd\u529b\u3002<\/span><\/p>\n


<\/section>\n


<\/p>\n

\n
\n
\n
\n

\u5927\u6a21\u578b2.0\u8bfb\u4e66\u4f1a\u542f\u52a8<\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n


<\/section>\n

<\/span><\/span><\/a><\/p>\n


<\/span><\/p>\n

\u8be6\u60c5\u8bf7\u89c1\uff1a\u5927\u6a21\u578b2.0\u8bfb\u4e66\u4f1a\uff1a\u878d\u5408\u5b66\u4e60\u4e0e\u63a8\u7406\u7684\u5927\u6a21\u578b\u65b0\u8303\u5f0f\uff01<\/a><\/span><\/p>\n


<\/section>\n
\n

<\/section>\n<\/section>\n
\u63a8\u8350\u9605\u8bfb<\/span><\/strong><\/strong><\/strong><\/section>\n
1. <\/strong>\u7c7b\u8111\u8ba1\u7b97\u6a21\u578b\u767bNature\u5b50\u520a\uff1a\u53d7\u5927\u8111\u542f\u53d1\u7684\u4eba\u5de5\u6811\u7a81\u7f51\u7edc\uff0c\u5b9e\u73b0\u9ad8\u80fd\u6548AI<\/a><\/strong><\/span><\/strong><\/section>\n
2. <\/strong>AI\u5b66\u4e60600\u4e07\u4eba\u751f\u6d3b\u4e8b\u4ef6\u5e8f\u5217\uff0c\u9884\u6d4b\u4eba\u7c7b\u751f\u6d3b\u8f68\u8ff9<\/a><\/strong><\/span><\/strong><\/section>\n
3. <\/strong>\u5317\u5927\u53d1\u8868 AI Alignment \u7efc\u8ff0\uff1a\u786e\u4fddAI\u4e0e\u4eba\u7c7b\u4ef7\u503c\u89c2\u4e00\u81f4\u7684\u56db\u4e2a\u5173\u952e\u8bbe\u8ba1\u539f\u5219<\/a><\/strong><\/strong><\/section>\n
4. <\/strong><\/span>\u5f20\u6c5f\uff1a\u7b2c\u4e09\u4ee3\u4eba\u5de5\u667a\u80fd\u6280\u672f\u57fa\u7840\u2014\u2014\u4ece\u53ef\u5fae\u5206\u7f16\u7a0b\u5230\u56e0\u679c\u63a8\u7406 | \u96c6\u667a\u5b66\u56ed\u5168\u65b0\u8bfe\u7a0b<\/a><\/strong><\/strong><\/span><\/section>\n
<\/strong>5. <\/strong>\u89e3\u9501\u96c6\u667a\u5168\u7ad9\u5185\u5bb9\uff0c\u5f00\u542f\u65b0\u5e74\u5b66\u4e60\u8ba1\u5212<\/a><\/strong><\/section>\n

6. <\/strong>\u52a0\u5165\u96c6\u667a\uff0c\u4e00\u8d77\u590d\u6742\uff01<\/strong><\/span><\/a><\/p>\n


<\/p>\n


<\/span><\/p>\n

\u70b9\u51fb\u201c\u9605\u8bfb\u539f\u6587\u201d\uff0c\u62a5\u540d\u8bfb\u4e66\u4f1a<\/span><\/section>\n

<\/mp-style-type><\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

\u5173\u952e\u8bcd\uff1a\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\uff0c\u76f4\u89c9\u7269\u7406\u5b66\uff0c\u56e0\u679c\u63a8\u7406\uff0c\u76f4\u89c9\u5fc3\u7406\u5b66\uff0c\u89c6\u89c9\u8ba4\u77e5 \u8bba\u6587\u9898\u76ee\uff1aVisual cognition in multimodal large language models \u8bba\u6587\u5730\u5740\uff1ahttps:\/\/www.nature.com\/articles\/s42256-024-00963-y \u671f\u520a…<\/p>\n","protected":false},"author":0,"featured_media":55395,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/55403"}],"collection":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=55403"}],"version-history":[{"count":0,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/55403\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/55395"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=55403"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=55403"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=55403"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=55403"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}