{"id":8950,"date":"2025-06-09T21:16:04","date_gmt":"2025-06-09T21:16:04","guid":{"rendered":"https:\/\/nationalgunowner.org\/index.php\/2025\/06\/09\/apple-research-finds-reasoning-a-i-models-arent-actually-reasoning\/"},"modified":"2025-06-09T21:16:05","modified_gmt":"2025-06-09T21:16:05","slug":"apple-research-finds-reasoning-a-i-models-arent-actually-reasoning","status":"publish","type":"post","link":"https:\/\/nationalgunowner.org\/index.php\/2025\/06\/09\/apple-research-finds-reasoning-a-i-models-arent-actually-reasoning\/","title":{"rendered":"Apple Research Finds \u2018Reasoning\u2019 A.I. Models Aren\u2019t Actually Reasoning"},"content":{"rendered":"<div itemprop=\"articleBody\">\n<figure id=\"attachment_1558918\" aria-describedby=\"caption-attachment-1558918\" style=\"width: 970px\" class=\"wp-caption aligncenter\"><figcaption id=\"caption-attachment-1558918\" class=\"wp-caption-text\">Compared with its Big Tech rivals, Apple\u2019s pace of A.I. development is cautiously slow. <span class=\"media-credit\">Justin Sullivan\/Getty Images<\/span><\/figcaption><\/figure>\n<p data-start=\"0\" data-end=\"666\">Just as the hype around artificial general intelligence (A.G.I.) reaches a fever pitch, <a href=\"https:\/\/observer.com\/company\/apple\/\" title=\"Apple\" class=\"company-link\">Apple<\/a> has delivered a sobering reality check to the industry. In a research paper titled \u201c<a target=\"_blank\" rel=\"noopener\" href=\"https:\/\/ml-site.cdn-apple.com\/papers\/the-illusion-of-thinking.pdf\" data-lasso-id=\"2776918\">The Illusion of Thinking<\/a>,\u201d published on June 6, the company argues that the most advanced A.I. models available today\u2014those billed as capable of \u201chuman-level reasoning\u201d\u2014struggle with complex logic problems. Instead of genuinely thinking like humans, these models rely on pattern recognition: drawing from familiar cues in their training data and predicting the next step. When faced with unfamiliar or challenging tasks, the models either offer weak responses or fail entirely.<\/p>\n<section class=\"wp-block-observer-newsletters observer-newsletters--in-content\">\n<\/section>\n<p data-start=\"668\" data-end=\"1223\">In a controlled study, Apple researchers tested large language models (LLMs) such as <a href=\"https:\/\/observer.com\/company\/anthropic\/\" title=\"Anthropic\" class=\"company-link\">Anthropic<\/a>\u2019s Claude 3.7 Sonnet, DeepSeek-V3, and their \u201creasoning-optimized\u201d versions (Claude 3.7 with Thinking and DeepSeek-R1). The team applied classic logic puzzles like the Tower of Hanoi and River Crossing\u2014well-established benchmarks for testing A.I. algorithms, planning and reasoning capabilities. The Tower of Hanoi tests recursion and step-by-step problem-solving, while River Crossing puzzles assess an A.I.\u2019s ability to plan and execute multi-step solutions.<\/p>\n<p data-start=\"1225\" data-end=\"1799\">Apple\u2019s researchers categorized the puzzles into three difficulty levels: low (3 steps), medium (4\u201310 steps) and high (11\u201320 steps). While most models handled the simpler tasks with reasonable success, their performance dropped dramatically as the puzzles grew more complex\u2014regardless of model size, training method or computational power. Even when given the correct algorithm or allowed to use up to 64,000 tokens\u2014a large computational budget\u2014the models offered only shallow responses, and performance did not improve even with explicit access to the solution algorithm.<\/p>\n<p data-start=\"1801\" data-end=\"2129\">Through this study, Apple researchers argue that what we often refer to as \u201creasoning\u201d may, in fact, be little more than sophisticated pattern-matching. They describe this phenomenon as a \u201ccounterintuitive scaling limit,\u201d where models, despite having ample computational resources, exert less effort as the complexity increases.<\/p>\n<p data-start=\"2131\" data-end=\"2567\">\u201cCurrent evaluations focus primarily on established mathematical and coding benchmarks, emphasizing final answer accuracy,\u201d Apple wrote in <a target=\"_blank\" rel=\"noopener\" href=\"https:\/\/machinelearning.apple.com\/research\/illusion-of-thinking\" data-lasso-id=\"2776919\">a blog post about the findings.<\/a> \u201cHowever, this paradigm often suffers from data contamination and fails to provide insights into the structure and quality of reasoning traces. Our setup allows analysis not only of the final answers but also of the internal reasoning traces, offering insights into how Large Reasoning Models (LRMs) \u2018think.\u2019\u201d<\/p>\n<p data-start=\"2569\" data-end=\"3203\">This study introduces much-needed rigor to a field often dominated by marketing hype, especially at a time when tech giants are touting A.G.I. as<a target=\"_blank\" rel=\"noopener\" href=\"https:\/\/blog.samaltman.com\/three-observations\" data-lasso-id=\"2776920\"> just around the corner.<\/a> It may also explain Apple\u2019s more cautious approach to A.I. development.<\/p>\n<h3 data-start=\"2569\" data-end=\"3203\">Apple reports its own A.I. progress at WWDC<\/h3>\n<p data-start=\"2569\" data-end=\"3203\">The research paper was dropped days before Apple\u2019s annual WWDC developers conference, which kicked off today.\u00a0In the opening keynote, Apple executives unveiled the Foundation Models framework. This framework will enable developers to integrate A.I. models into their apps, facilitating capabilities such as image generation, text creation and natural language search.<\/p>\n<article class=\"text-token-text-primary w-full\" dir=\"auto\" data-testid=\"conversation-turn-2\" data-scroll-anchor=\"true\">\n<div class=\"text-base my-auto mx-auto py-5 [--thread-content-margin:--spacing(4)] @[37rem]:[--thread-content-margin:--spacing(6)] @[72rem]:[--thread-content-margin:--spacing(16)] px-(--thread-content-margin)\">\n<div class=\"[--thread-content-max-width:32rem] @[34rem]:[--thread-content-max-width:40rem] @[64rem]:[--thread-content-max-width:48rem] mx-auto flex max-w-(--thread-content-max-width) flex-1 text-base gap-4 md:gap-5 lg:gap-6 group\/turn-messages focus-visible:outline-hidden\">\n<div class=\"group\/conversation-turn relative flex w-full min-w-0 flex-col agent-turn\">\n<div class=\"relative flex-col gap-1 md:gap-3\">\n<div class=\"flex max-w-full flex-col grow\">\n<div class=\"min-h-8 text-message relative flex w-full flex-col items-end gap-2 text-start break-words whitespace-normal [.text-message+&amp;]:mt-5\" dir=\"auto\" data-message-author-role=\"assistant\" data-message-id=\"1c863ccc-dd3d-4a5e-8a4b-6f2e2db715c5\" data-message-model-slug=\"gpt-4o-mini\">\n<div class=\"flex w-full flex-col gap-1 empty:hidden first:pt-[3px]\">\n<div class=\"markdown prose dark:prose-invert w-full break-words light\">\n<p data-start=\"3568\" data-end=\"4069\" data-is-last-node=\"\" data-is-only-node=\"\">Apple also introduced Xcode 26, a major update to its developer toolkit, which now includes built-in support for integrating A.I. models like ChatGPT and Claude via API keys. This update allows developers to leverage A.I. models for tasks like writing code, generating tests and documentation, and debugging. Together, these announcements mark a significant step in Apple\u2019s A.I. strategy, aiming to empower developers to build intelligent applications without relying on cloud infrastructure.<\/p>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/article>\n<p>\t\t\t\t<img decoding=\"async\" itemprop=\"image\" src=\"https:\/\/observer.com\/wp-content\/uploads\/sites\/2\/2025\/06\/GettyImages-1496169958.jpg?quality=80&amp;w=970\" alt=\"Apple Research Finds \u2018Reasoning\u2019 A.I. Models Aren\u2019t Actually Reasoning\" style=\"display:none;width:0;\"\/><\/p><\/div>\n<p><script>\n\t!function(f,b,e,v,n,t,s)\n\t{if(f.fbq)return;n=f.fbq=function(){n.callMethod?\n\t\tn.callMethod.apply(n,arguments):n.queue.push(arguments)};\n\t\tif(!f._fbq)f._fbq=n;n.push=n;n.loaded=!0;n.version='2.0';\n\t\tn.queue=[];t=b.createElement(e);t.async=!0;\n\t\tt.src=v;s=b.getElementsByTagName(e)[0];\n\t\ts.parentNode.insertBefore(t,s)}(window, document,'script',\n\t\t'https:\/\/connect.facebook.net\/en_US\/fbevents.js');\n\tfbq('init', '618909876214345');\n\tfbq('track', 'PageView');\n<\/script><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Compared with its Big Tech rivals, Apple\u2019s pace of A.I. development is cautiously slow. Justin Sullivan\/Getty Images Just as the hype around artificial general intelligence (A.G.I.) reaches a fever pitch, Apple has delivered a sobering reality check to the industry. In a research paper titled \u201cThe Illusion of Thinking,\u201d published on June 6, the company [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":8951,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[10],"tags":[],"class_list":["post-8950","post","type-post","status-publish","format-standard","has-post-thumbnail","category-usa-news"],"_links":{"self":[{"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/posts\/8950","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/comments?post=8950"}],"version-history":[{"count":1,"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/posts\/8950\/revisions"}],"predecessor-version":[{"id":8952,"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/posts\/8950\/revisions\/8952"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/media\/8951"}],"wp:attachment":[{"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/media?parent=8950"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/categories?post=8950"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/nationalgunowner.org\/index.php\/wp-json\/wp\/v2\/tags?post=8950"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}