{"id":37016,"date":"2025-06-08T18:35:45","date_gmt":"2025-06-08T10:35:45","guid":{"rendered":"https:\/\/www.1ai.net\/?p=37016"},"modified":"2025-06-08T18:35:45","modified_gmt":"2025-06-08T10:35:45","slug":"%e8%8b%b9%e6%9e%9c%e6%9c%80%e6%96%b0%e7%a0%94%e7%a9%b6%ef%bc%9a%e7%8e%b0%e6%9c%89-ai-%e5%a4%a7%e6%a8%a1%e5%9e%8b%e6%9b%b4%e5%83%8f%e6%98%af%e5%9c%a8%e8%ae%b0%e5%bf%86%ef%bc%8c%e8%80%8c","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/37016.html","title":{"rendered":"New Apple study: existing big AI models are \"more like memorization than real reasoning\""},"content":{"rendered":"<p>6 \u6708 8 \u65e5\u6d88\u606f\uff0c<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e8%8b%b9%e6%9e%9c\" title=\"\u3010\u67e5\u770b\u542b\u6709[\u82f9\u679c]\u6807\u7b7e\u7684\u6587\u7ae0\u3011\" target=\"_blank\" rel=\"noopener\">\u82f9\u679c<\/a>\u673a\u5668\u5b66\u4e60\u7814\u7a76\u4e2d\u5fc3\u4e8e\u5f53\u5730\u65f6\u95f4 6 \u6708 6 \u65e5\u53d1\u8868\u4e86\u4e00\u7bc7<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e7%a0%94%e7%a9%b6%e8%ae%ba%e6%96%87\" title=\"\u3010\u67e5\u770b\u542b\u6709[\u7814\u7a76\u8bba\u6587]\u6807\u7b7e\u7684\u6587\u7ae0\u3011\" target=\"_blank\" rel=\"noopener\">\u7814\u7a76\u8bba\u6587<\/a>\uff0c\u79f0\u73b0\u6709 <a href=\"https:\/\/www.1ai.net\/en\/tag\/ai%e6%a8%a1%e5%9e%8b\" title=\"\u3010\u67e5\u770b\u542b\u6709[AI\u6a21\u578b]\u6807\u7b7e\u7684\u6587\u7ae0\u3011\" target=\"_blank\" rel=\"noopener\">AI\u6a21\u578b<\/a>\u5e76\u4e0d\u5177\u5907\u771f\u6b63\u7684\u601d\u7ef4\u80fd\u529b\u6216\u63a8\u7406\u80fd\u529b\uff0c\u800c\u662f\u4f9d\u8d56\u4e8e\u6a21\u5f0f\u5339\u914d\u4e0e\u8bb0\u5fc6\uff0c\u5c24\u5176\u662f\u5bf9\u4e8e\u590d\u6742\u7684\u4efb\u52a1\u800c\u8a00\u3002<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-37017\" title=\"6507e202j00sxj9dq002hd000la00fap\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2025\/06\/6507e202j00sxj9dq002hd000la00fap.jpg\" alt=\"6507e202j00sxj9dq002hd000la00fap\" width=\"766\" height=\"550\" \/><\/p>\n<p>\u82f9\u679c\u7814\u7a76\u4eba\u5458\u5bf9\u73b0\u6709\u7684\u524d\u6cbf\u201c\u5927\u578b\u63a8\u7406\u6a21\u578b\u201d\u2014\u2014 \u5982 OpenAI o3-mini\u3001DeepSeek-R1\u3001Anthropic \u7684 Claude 3.7 Sonnet Thinking \u548c\u8c37\u6b4c Gemini Thinking\u2014\u2014 \u8fdb\u884c\u4e86\u7cfb\u7edf\u8bc4\u4f30\u3002<\/p>\n<p>\u7814\u7a76\u53d1\u73b0\uff0c\u5c3d\u7ba1\u8fd9\u4e9b\u6a21\u578b\u5177\u5907\u751f\u6210\u8be6\u7ec6\u201c\u601d\u8003\u94fe\u201d\u7684\u80fd\u529b\uff0c\u5e76\u5728\u4e2d\u7b49\u590d\u6742\u5ea6\u4efb\u52a1\u4e0a\u8868\u73b0\u51fa\u4f18\u52bf\uff0c\u4f46\u5176\u63a8\u7406\u80fd\u529b\u5b58\u5728\u6839\u672c\u6027\u5c40\u9650\uff1a\u5f53\u95ee\u9898\u590d\u6742\u5ea6\u8d85\u8fc7\u7279\u5b9a\u4e34\u754c\u70b9\u65f6\uff0c\u6a21\u578b\u6027\u80fd\u4f1a\u5b8c\u5168\u5d29\u6e83\u81f3\u201c\u96f6\u51c6\u786e\u7387\u201d\u3002<\/p>\n<p>\u6b64\u5916\uff0c\u5728\u6a21\u578b\u63a8\u7406\u8fc7\u7a0b\u4e2d\uff0c\u5373\u4f7f\u4ecd\u6709\u5145\u8db3\u7684\u63a8\u7406\u7b97\u529b\uff0c\u5b83\u4eec\u7528\u4e8e\u201c\u601d\u8003\u201d\u7684 token \u6570\u91cf\u53cd\u800c\u968f\u96be\u5ea6\u4e0a\u5347\u800c\u51cf\u5c11\uff0c\u8fd9\u79cd\u73b0\u8c61\u610f\u5473\u7740\u73b0\u6709\u63a8\u7406\u65b9\u6cd5\u5b58\u5728\u6839\u672c\u5c40\u9650\u6027\u3002<\/p>\n<p>\u8fd9\u7bc7\u300a\u601d\u8003\u7684\u5e7b\u8c61\uff1a\u901a\u8fc7\u95ee\u9898\u590d\u6742\u6027\u7684\u89c6\u89d2\u7406\u89e3\u63a8\u7406\u6a21\u578b\u7684\u4f18\u52bf\u4e0e\u5c40\u9650\u300b\u7531 Parshin Shojaee \u7b49\u4eba\u64b0\u5199\u3002\u7814\u7a76\u8868\u660e\uff0c\u5f53\u524d\u4e1a\u754c\u5bf9\u8fd9\u4e9b\u6a21\u578b\u7684\u8bc4\u4f30\u4e3b\u8981\u96c6\u4e2d\u5728\u6570\u5b66\u548c\u7f16\u7a0b\u57fa\u51c6\u6d4b\u8bd5\u4e0a\uff0c\u5173\u6ce8\u6700\u7ec8\u7b54\u6848\u7684\u51c6\u786e\u6027\uff0c\u4f46\u8fd9\u5f80\u5f80\u5ffd\u7565\u4e86\u6570\u636e\u6c61\u67d3\u95ee\u9898\uff0c\u4e5f\u65e0\u6cd5\u63d0\u4f9b\u6709\u5173\u5185\u90e8\u63a8\u7406\u8f68\u8ff9\u7ed3\u6784\u548c\u8d28\u91cf\u7684\u6d1e\u89c1\u3002<\/p>\n<p>\u7814\u7a76\u4eba\u5458\u91c7\u7528\u4e86\u4e00\u7cfb\u5217\u53ef\u63a7\u7684\u89e3\u8c1c\u73af\u5883\uff0c\u5141\u8bb8\u7cbe\u786e\u64cd\u7eb5\u7ec4\u6210\u590d\u6742\u6027\uff0c\u540c\u65f6\u4fdd\u6301\u903b\u8f91\u7ed3\u6784\u7684\u4e00\u81f4\u6027\u3002\u8fd9\u4f7f\u5f97\u4e0d\u4ec5\u53ef\u4ee5\u5206\u6790\u6700\u7ec8\u7b54\u6848\uff0c\u8fd8\u53ef\u4ee5\u63a2\u7a76\u5185\u90e8\u63a8\u7406\u8f68\u8ff9\uff0c\u4ece\u800c\u66f4\u6df1\u5165\u5730\u4e86\u89e3\u8fd9\u4e9b\u6a21\u578b\u662f\u5982\u4f55\u201c\u601d\u8003\u201d\u7684\u3002<\/p>\n<p>\u7814\u7a76\u56e2\u961f\u63d0\u51fa\uff0c\u6a21\u578b\u8868\u73b0\u53ef\u5206\u4e3a\u4e09\u4e2a\u9636\u6bb5\uff1a<\/p>\n<ul>\n<li><strong>\u4f4e\u590d\u6742\u5ea6\u4efb\u52a1<\/strong>\uff1a\u4f20\u7edf\u5927\u6a21\u578b\uff08\u6ce8\uff1a\u5982 Claude-3.7 \u65e0\u601d\u7ef4\u7248\u672c\uff09\u8868\u73b0\u66f4\u4f73\uff1b<\/li>\n<li><strong>\u4e2d\u7b49\u590d\u6742\u5ea6\u4efb\u52a1<\/strong>\uff1a\u5177\u5907\u601d\u7ef4\u673a\u5236\u7684\u5927\u578b\u63a8\u7406\u6a21\u578b\uff08LRMs\uff09\u66f4\u5360\u4f18\u52bf\uff1b<\/li>\n<li><strong>\u9ad8\u590d\u6742\u5ea6\u4efb\u52a1<\/strong>\uff1a\u4e24\u7c7b\u6a21\u578b\u5747\u9677\u5165\u5b8c\u5168\u5931\u6548\u72b6\u6001\u3002<\/li>\n<\/ul>\n<p>\u7279\u522b\u662f\uff0c\u7814\u7a76\u53d1\u73b0 LRMs \u5728\u6267\u884c\u7cbe\u786e\u8ba1\u7b97\u65b9\u9762\u5b58\u5728\u5c40\u9650\u6027\uff0c\u65e0\u6cd5\u4f7f\u7528\u663e\u5f0f\u7b97\u6cd5\u4e14\u8de8\u4e0d\u540c\u8c1c\u9898\u8fdb\u884c\u63a8\u7406\u65f6\u8868\u73b0\u51fa\u4e0d\u4e00\u81f4\u6027\u3002<\/p>\n<p>\u603b\u7684\u6765\u8bf4\uff0c\u8fd9\u9879\u7814\u7a76\u4e0d\u4ec5\u8d28\u7591\u4e86\u5f53\u524d\u57fa\u4e8e\u5df2\u5efa\u7acb\u6570\u5b66\u57fa\u51c6\u7684 LRMs \u8bc4\u4f30\u8303\u5f0f\uff0c\u8fd8\u5f3a\u8c03\u4e86\u9700\u8981\u66f4\u52a0\u7ec6\u81f4\u7684\u5b9e\u9a8c\u8bbe\u7f6e\u6765\u63a2\u7d22\u8fd9\u4e9b\u95ee\u9898\u3002\u901a\u8fc7\u4f7f\u7528\u53ef\u63a7\u5236\u7684\u8c1c\u9898\u73af\u5883\uff0c\u672c\u7814\u7a76\u63d0\u4f9b\u4e86\u5bf9\u8bed\u8a00\u63a8\u7406\u6a21\u578b\u80fd\u529b\u548c\u5c40\u9650\u6027\u7684\u6df1\u523b\u89c1\u89e3\uff0c\u5e76\u4e3a\u672a\u6765\u7684\u7814\u7a76\u6307\u660e\u4e86\u65b9\u5411\u3002<\/p>\n<p>\u7814\u7a76\u4eba\u5458\u8868\u793a\uff0c\u201c\u8fd9\u4e9b\u53d1\u73b0\u7a81\u51fa\u4e86\u73b0\u6709 LRMs \u7684\u4f18\u70b9\u548c\u5c40\u9650\u6027\uff0c\u5f15\u53d1\u4e86\u5173\u4e8e\u8fd9\u4e9b\u7cfb\u7edf\u63a8\u7406\u672c\u8d28\u7684\u95ee\u9898\uff0c\u8fd9\u5bf9\u5b83\u4eec\u7684\u8bbe\u8ba1\u548c\u90e8\u7f72\u5177\u6709\u91cd\u8981\u610f\u4e49\u3002\u201d<\/p>\n<p data-vmark=\"43dd\"><span class=\"referenceTitle\">\u53c2\u8003\u8d44\u6599\uff1a<\/span><\/p>\n<ul class=\"custom_reference list-paddingleft-1\">\n<li class=\"list-undefined list-reference-paddingleft\">\n<p data-vmark=\"ed50\">\u300a<a href=\"https:\/\/machinelearning.apple.com\/research\/illusion-of-thinking\" target=\"_blank\" rel=\"noopener\">The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity &#8211; Apple Machine Learning Research<\/a>\u300b<\/p>\n<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>June 8, 2011 - Apple's Machine Learning Research Center published a research paper on June 6, local time, claiming that existing AI models don't really have the ability to think or reason, but instead rely on pattern matching and memory, especially for complex tasks. The Apple researchers examined existing cutting-edge \"large-scale inference models\" - such as OpenAI o3-mini, DeepSeek-R1, Anthropic's Claude 3.7 Sonnet Thinking Anthropic's Claude 3.7 Sonnet Thinking, and Google Gemini Thinking - were systematically evaluated. The study found that while these models have the ability to generate detailed \"thought chains\" and show strengths on medium-complexity tasks, they are not as good at generating detailed \"thought chains\" as they are at generating detailed \"thought chains\".<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[146],"tags":[167,2760,345],"collection":[],"class_list":["post-37016","post","type-post","status-publish","format-standard","hentry","category-news","tag-ai","tag-2760","tag-345"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/37016","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=37016"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/37016\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=37016"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=37016"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=37016"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=37016"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}