{"id":1136,"date":"2026-08-13T00:30:26","date_gmt":"2026-08-12T15:30:26","guid":{"rendered":"https:\/\/itexplore.org\/jp\/columns\/ai-cyber-defense-math-discoveries-inner-thoughts\/"},"modified":"2026-08-13T00:30:26","modified_gmt":"2026-08-12T15:30:26","slug":"ai-cyber-defense-math-discoveries-inner-thoughts","status":"publish","type":"post","link":"https:\/\/itexplore.org\/jp\/columns\/ai-cyber-defense-math-discoveries-inner-thoughts\/","title":{"rendered":"AI\u306e\u30b5\u30a4\u30d0\u30fc\u9632\u5fa1\u3068\u6570\u5b66\u7684\u767a\u898b\u3001\u305d\u3057\u3066\u5185\u90e8\u601d\u8003\u306e\u89e3\u660e"},"content":{"rendered":"<p>\u672c\u65e5\u306e\u6ce8\u76eeAI\u30fb\u30c6\u30c3\u30af\u30cb\u30e5\u30fc\u30b9\u3092\u3001\u5c02\u9580\u7684\u306a\u5206\u6790\u3068\u5171\u306b\u304a\u5c4a\u3051\u3057\u307e\u3059\u3002<\/p>\n<div class=\"wp-block-vk-blocks-alert vk_alert alert alert-warning has-alert-icon\">\n<div class=\"vk_alert_icon\">\n<div class=\"vk_alert_icon_icon\"><i class=\"fa-solid fa-triangle-exclamation\" aria-hidden=\"true\"><\/i><\/div>\n<div class=\"vk_alert_icon_text\"><span>Warning<\/span><\/div>\n<\/div>\n<div class=\"vk_alert_content\">\n<p>\u3053\u306e\u8a18\u4e8b\u306fAI\u306b\u3088\u3063\u3066\u81ea\u52d5\u751f\u6210\u30fb\u5206\u6790\u3055\u308c\u305f\u3082\u306e\u3067\u3059\u3002AI\u306e\u6027\u8cea\u4e0a\u3001\u4e8b\u5b9f\u8aa4\u8a8d\u304c\u542b\u307e\u308c\u308b\u53ef\u80fd\u6027\u304c\u3042\u308b\u305f\u3081\u3001\u91cd\u8981\u306a\u5224\u65ad\u3092\u4e0b\u3059\u969b\u306f\u5fc5\u305a\u30ea\u30f3\u30af\u5148\u306e\u4e00\u6b21\u30bd\u30fc\u30b9\u3092\u3054\u78ba\u8a8d\u304f\u3060\u3055\u3044\u3002<\/p>\n<\/div>\n<\/div>\n<div class=\"wp-block-group\" style=\"margin-top:40px;margin-bottom:40px\">\n<h2 class=\"wp-block-heading\">\u30b5\u30a4\u30d0\u30fc\u9632\u5fa1\u306e\u7a93\u304c\u72ed\u307e\u308b\u4e2d\u3001Daybreak\u3092\u62e1\u5927<\/h2>\n<ul>\n<li><strong>\u539f\u984c:<\/strong> Expanding Daybreak as the Cyber Defense Window Narrows | OpenAI<\/li>\n<\/ul>\n<h3 class=\"wp-block-heading\">\u5c02\u9580\u30a2\u30ca\u30ea\u30b9\u30c8\u306e\u5206\u6790<\/h3>\n<div class=\"ai-summary-content\">\n<p><strong>OpenAI<\/strong>\u306f\u3001\u30b5\u30a4\u30d0\u30fc\u653b\u6483\u8005\u304cAI\u3092\u60aa\u7528\u3059\u308b\u901f\u5ea6\u3068\u898f\u6a21\u304c\u5897\u5927\u3059\u308b\u4e2d\u3001\u9632\u5fa1\u5074\u306e\u6e96\u5099\u671f\u9593\u304c\u72ed\u307e\u3063\u3066\u3044\u308b\u73fe\u72b6\u306b\u5bfe\u5fdc\u3059\u308b\u305f\u3081\u3001\u30b5\u30a4\u30d0\u30fc\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u306b\u7279\u5316\u3057\u305f\u6700\u65b0\u30e2\u30c7\u30eb<strong>GPT-5.6-Cyber<\/strong>\u3068<strong>Daybreak<\/strong>\u30d7\u30ed\u30b0\u30e9\u30e0\u3092\u767a\u8868\u3057\u307e\u3057\u305f\u3002\u3053\u306e\u30d7\u30ed\u30b0\u30e9\u30e0\u306f\u3001\u627f\u8a8d\u3055\u308c\u305f\u9632\u5fa1\u8005\u306b\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2AI\u306e\u80fd\u529b\u3092\u63d0\u4f9b\u3059\u308b\u3053\u3068\u3092\u76ee\u7684\u3068\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<p><strong>Daybreak<\/strong>\u306b\u306f\u4e8c\u3064\u306e\u30a2\u30af\u30bb\u30b9\u5c64\u304c\u3042\u308a\u307e\u3059\u3002<strong>Daybreak Blue<\/strong>\u306f\u3001<strong>GPT-5.6 Sol<\/strong>\u3092\u542b\u3080\u6c4e\u7528\u30e2\u30c7\u30eb\u3078\u306e\u30a2\u30af\u30bb\u30b9\u3092\u63d0\u4f9b\u3057\u3001\u8106\u5f31\u6027\u767a\u898b\u3001\u30bb\u30ad\u30e5\u30a2\u30b3\u30fc\u30c9\u30ec\u30d3\u30e5\u30fc\u3001\u30de\u30eb\u30a6\u30a7\u30a2\u5206\u6790\u306a\u3069\u306e\u9632\u5fa1\u7684\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u4f5c\u696d\u306b\u7279\u5316\u3057\u305f\u30bb\u30fc\u30d5\u30ac\u30fc\u30c9\u3092\u5099\u3048\u3066\u3044\u307e\u3059\u3002\u4e00\u65b9\u3001<strong>Daybreak Red<\/strong>\u306f\u3001<strong>GPT-5.6-Cyber<\/strong>\u306e\u3088\u3046\u306a\u76ee\u7684\u5225\u306b\u8a13\u7df4\u3055\u308c\u305f\u30b5\u30a4\u30d0\u30fc\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u30e2\u30c7\u30eb\u3078\u306e\u30a2\u30af\u30bb\u30b9\u3092\u63d0\u4f9b\u3057\u3001\u8106\u5f31\u6027\u7814\u7a76\u3084\u30a8\u30af\u30b9\u30d7\u30ed\u30a4\u30c8\u691c\u8a3c\u3001\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u30c6\u30b9\u30c8\u3092\u652f\u63f4\u3057\u307e\u3059\u3002<\/p>\n<p><strong>GPT-5.6-Cyber<\/strong>\u306f\u3001<strong>GPT-5.6 Sol<\/strong>\u3092\u57fa\u76e4\u3068\u3057\u3001\u30bc\u30ed\u30c7\u30a4\u8106\u5f31\u6027\u306e\u767a\u898b\u3084\u30a8\u30af\u30b9\u30d7\u30ed\u30a4\u30c8\u30c1\u30a7\u30fc\u30f3\u306e\u958b\u767a\u3068\u3044\u3063\u305f\u5c02\u9580\u7684\u306a\u30b5\u30a4\u30d0\u30fc\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u30bf\u30b9\u30af\u306b\u304a\u3044\u3066\u80fd\u529b\u3092\u5411\u4e0a\u3055\u305b\u3001\u7279\u5b9a\u306e\u9ad8\u30ea\u30b9\u30af\u306a\u30b5\u30a4\u30d0\u30fc\u30bf\u30b9\u30af\u306b\u304a\u3051\u308b\u62d2\u5426\u7387\u3092\u5927\u5e45\u306b\u524a\u6e1b\u3057\u307e\u3059\u3002\u5185\u90e8\u8a55\u4fa1\u3067\u306f\u3001<strong>GPT-5.6-Cyber<\/strong>\u304c\u9ad8\u5ea6\u306a\u30b5\u30a4\u30d0\u30fc\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u8981\u6c42\u306e95.0%\u3092\u5b8c\u4e86\u3059\u308b\u306e\u306b\u5bfe\u3057\u3001<strong>GPT-5.6 Sol<\/strong>\u306f1.5%\u306b\u7559\u307e\u308a\u307e\u3059\u3002<\/p>\n<p>\u3053\u306e\u30e2\u30c7\u30eb\u306f\u3001<strong>V8<\/strong> JavaScript\u30a8\u30f3\u30b8\u30f3\u306b\u304a\u3051\u308b2\u3064\u306e\u672a\u77e5\u306e\u8106\u5f31\u6027\uff08<strong>CVE-2026-15903<\/strong>\uff09\u3092\u767a\u898b\u3057\u3001Google\u306b\u5831\u544a\u3059\u308b\u306a\u3069\u3001\u5b9f\u969b\u306e\u8106\u5f31\u6027\u7814\u7a76\u3067\u305d\u306e\u6709\u52b9\u6027\u3092\u793a\u3057\u307e\u3057\u305f\u3002\u307e\u305f\u3001\u4eba\u6c17\u306e\u3042\u308b\u30e2\u30d0\u30a4\u30ebOS\u3001\u30c7\u30fc\u30bf\u30d9\u30fc\u30b9\u3001OS\u30ab\u30fc\u30cd\u30eb\u306b\u304a\u3044\u3066\u3001\u5408\u8a08\u3067400\u4ee5\u4e0a\u306e\u9ad8\u6df1\u523b\u5ea6\u306a\u8106\u5f31\u6027\u3092\u7279\u5b9a\u3057\u3066\u3044\u307e\u3059\u3002<strong>OpenAI<\/strong>\u306f\u3001\u3053\u308c\u3089\u306e\u30e2\u30c7\u30eb\u306e\u5b89\u5168\u306a\u5229\u7528\u3092\u78ba\u4fdd\u3059\u308b\u305f\u3081\u3001\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u30bb\u30ad\u30e5\u30ea\u30c6\u30a3\u30ad\u30fc\u306e\u7fa9\u52d9\u4ed8\u3051\u3084\u76e3\u8996\u5f37\u5316\u306a\u3069\u306e\u8ffd\u52a0\u63aa\u7f6e\u3092\u8b1b\u3058\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\ud83d\udc49 <strong><a href=\"https:\/\/openai.com\/index\/expanding-daybreak-as-the-cyber-defense-window-narrows\/\" target=\"_blank\" rel=\"noopener\">OpenAI \u3067\u8a18\u4e8b\u5168\u6587\u3092\u8aad\u3080<\/a><\/strong><\/p>\n<\/div>\n<ul>\n<li><strong>\u8981\u70b9:<\/strong> OpenAI's GPT-5.6-Cyber and Daybreak program significantly enhance cyber defense capabilities by providing specialized AI models with reduced refusal rates for authorized security tasks, leading to the discovery of critical real-world vulnerabilities.<\/li>\n<li><strong>\u8457\u8005:<\/strong> OpenAI<\/li>\n<\/ul>\n<blockquote class=\"wp-block-quote\"><p><span>English Summary:<\/span><\/p>\n<p><strong>OpenAI<\/strong> has introduced its latest cybersecurity-specific model, <strong>GPT-5.6-Cyber<\/strong>, and the <strong>Daybreak<\/strong> program, in response to the rapidly narrowing window for cyber defense as threat actors increasingly leverage AI for attacks. The program aims to equip approved defenders with frontier AI capabilities.<\/p>\n<p><strong>Daybreak<\/strong> offers two access tiers: <strong>Daybreak Blue<\/strong> provides access to frontier general-purpose models, including <strong>GPT-5.6 Sol<\/strong>, with safeguards tailored for authorized defensive security work such as vulnerability discovery and secure code review. <strong>Daybreak Red<\/strong> offers access to purpose-trained cybersecurity models like <strong>GPT-5.6-Cyber<\/strong> for authorized vulnerability research and exploit validation.<\/p>\n<p><strong>GPT-5.6-Cyber<\/strong>, built on <strong>GPT-5.6 Sol<\/strong>, is specifically trained to enhance capabilities in specialized cybersecurity tasks, such as finding zero-day vulnerabilities and developing exploit chains, while significantly reducing refusals for certain higher-risk cyber tasks. Internal evaluations show <strong>GPT-5.6-Cyber<\/strong> completing 95.0% of advanced cybersecurity requests, compared to just 1.5% for <strong>GPT-5.6 Sol<\/strong>.<\/p>\n<p>The model has demonstrated its effectiveness in real-world vulnerability research, uncovering two previously unknown vulnerabilities in the <strong>V8<\/strong> JavaScript engine (<strong>CVE-2026-15903<\/strong>) and reporting them to Google. It has also identified over 400 high-severity issues across popular mobile operating systems, databases, and OS kernels. <strong>OpenAI<\/strong> is implementing additional safeguards, including mandatory hardware security keys and enhanced monitoring, to ensure the safe use of these models.<\/p>\n<\/blockquote>\n<\/div>\n<div class=\"wp-block-group\" style=\"margin-top:40px;margin-bottom:40px\">\n<h2 class=\"wp-block-heading\">AI\u30e2\u30c7\u30eb\u306e\u5185\u90e8\u601d\u8003\u3092\u660e\u3089\u304b\u306b\u3059\u308b\u65b0\u3057\u3044\u624b\u6cd5<\/h2>\n<ul>\n<li><strong>\u539f\u984c:<\/strong> A New Trick Reveals AI Models\u2019 Inner Thoughts<\/li>\n<\/ul>\n<h3 class=\"wp-block-heading\">\u5c02\u9580\u30a2\u30ca\u30ea\u30b9\u30c8\u306e\u5206\u6790<\/h3>\n<div class=\"ai-summary-content\">\n<p><strong>Will Knight<\/strong>\u306b\u3088\u308b\u8a18\u4e8b\u300cA New Trick Reveals AI Models\u2019 Inner Thoughts\u300d\u306f\u3001AI\u30e2\u30c7\u30eb\u3001\u7279\u306b\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb\uff08<strong>LLM<\/strong>\uff09\u304c\u3069\u306e\u3088\u3046\u306b\u610f\u601d\u6c7a\u5b9a\u3092\u884c\u3044\u3001\u51fa\u529b\u3092\u751f\u6210\u3059\u308b\u306e\u304b\u3068\u3044\u3046\u3001\u3044\u308f\u3086\u308b\u300c\u30d6\u30e9\u30c3\u30af\u30dc\u30c3\u30af\u30b9\u300d\u554f\u984c\u306b\u53d6\u308a\u7d44\u3080\u65b0\u3057\u3044\u7814\u7a76\u3084\u6280\u8853\u306b\u7126\u70b9\u3092\u5f53\u3066\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\u3053\u306e\u300c\u65b0\u3057\u3044\u624b\u6cd5\u300d\u306f\u3001AI\u306e\u5185\u90e8\u8868\u73fe\u3092\u8abf\u67fb\u3057\u3001\u305d\u306e\u63a8\u8ad6\u30d7\u30ed\u30bb\u30b9\u3084\u610f\u601d\u6c7a\u5b9a\u7d4c\u8def\u3092\u53ef\u8996\u5316\u3059\u308b\u3053\u3068\u3067\u3001\u30e2\u30c7\u30eb\u304c\u3069\u306e\u3088\u3046\u306b\u7d50\u8ad6\u306b\u81f3\u308b\u306e\u304b\u3092\u3088\u308a\u6df1\u304f\u7406\u89e3\u3059\u308b\u3053\u3068\u3092\u53ef\u80fd\u306b\u3059\u308b\u3082\u306e\u3067\u3059\u3002\u3053\u308c\u306b\u3088\u308a\u3001AI\u306e\u4fe1\u983c\u6027\u3001\u8aac\u660e\u53ef\u80fd\u6027\u3001\u305d\u3057\u3066\u30c7\u30d0\u30c3\u30b0\u80fd\u529b\u304c\u5411\u4e0a\u3059\u308b\u3053\u3068\u304c\u671f\u5f85\u3055\u308c\u307e\u3059\u3002<\/p>\n<p>\u8a18\u4e8b\u306f\u3001AI\u306e\u300c\u601d\u8003\u300d\u3092\u89e3\u660e\u3059\u308b\u305f\u3081\u306e\u9769\u65b0\u7684\u306a\u30a2\u30d7\u30ed\u30fc\u30c1\u304c\u3001AI\u7814\u7a76\u306b\u304a\u3051\u308b\u91cd\u8981\u306a\u9032\u6b69\u3092\u793a\u3057\u3066\u3044\u308b\u3053\u3068\u3092\u793a\u5506\u3057\u3066\u304a\u308a\u3001AI\u306e\u900f\u660e\u6027\u3068\u7406\u89e3\u3092\u6df1\u3081\u308b\u305f\u3081\u306e\u65b0\u305f\u306a\u9053\u3092\u958b\u304f\u3082\u306e\u3067\u3059\u3002<\/p>\n<p>\ud83d\udc49 <strong><a href=\"https:\/\/www.wired.com\/story\/a-new-trick-reveals-ai-models-inner-thoughts\/\" target=\"_blank\" rel=\"noopener\">Wired \u3067\u8a18\u4e8b\u5168\u6587\u3092\u8aad\u3080<\/a><\/strong><\/p>\n<\/div>\n<ul>\n<li><strong>\u8981\u70b9:<\/strong> New techniques are emerging to reveal the internal workings and 'thoughts' of AI models, particularly LLMs, enhancing interpretability, trustworthiness, and explainability in AI systems.<\/li>\n<li><strong>\u8457\u8005:<\/strong> Will Knight<\/li>\n<\/ul>\n<blockquote class=\"wp-block-quote\"><p><span>English Summary:<\/span><\/p>\n<p>The article by <strong>Will Knight<\/strong>, titled \"A New Trick Reveals AI Models\u2019 Inner Thoughts,\" focuses on novel research and techniques addressing the \"black box\" problem of how AI models, particularly <strong>Large Language Models (LLMs)<\/strong>, make decisions and generate outputs.<\/p>\n<p>This \"new trick\" likely involves innovative methods to probe the internal representations of AI, visualize their reasoning processes, and identify decision-making pathways, thereby offering deeper insights into how models arrive at their conclusions. Such advancements are expected to enhance AI trustworthiness, explainability, and debugging capabilities.<\/p>\n<p>The article suggests that these groundbreaking approaches to unveiling AI's \"thoughts\" represent a significant step forward in AI research, paving the way for greater transparency and understanding of artificial intelligence.<\/p>\n<\/blockquote>\n<\/div>\n<div class=\"wp-block-group\" style=\"margin-top:40px;margin-bottom:40px\">\n<h2 class=\"wp-block-heading\">Claude\u306e\u6570\u5b66\u7684\u80fd\u529b\u306b\u95a2\u3059\u308b\u3055\u3089\u306a\u308b\u5b66\u7fd2<\/h2>\n<ul>\n<li><strong>\u539f\u984c:<\/strong> Learning more about Claude's mathematical capabilities<\/li>\n<\/ul>\n<h3 class=\"wp-block-heading\">\u5c02\u9580\u30a2\u30ca\u30ea\u30b9\u30c8\u306e\u5206\u6790<\/h3>\n<div class=\"ai-summary-content\">\n<p><strong>Anthropic<\/strong>\u306e\u7814\u7a76\u8a18\u4e8b\u306f\u3001\u672a\u516c\u958b\u306e\u7814\u7a76\u7248<strong>Claude<\/strong>\u304c\u3001\u6570\u5b66\u306b\u304a\u3051\u308b\u6700\u3082\u6709\u540d\u306a\u672a\u89e3\u6c7a\u554f\u984c\u306e\u4e00\u3064\u3067\u3042\u308b<strong>\u30ea\u30fc\u30de\u30f3\u4e88\u60f3<\/strong>\u306b\u95a2\u9023\u3059\u308b\u753b\u671f\u7684\u306a\u9032\u5c55\u3092\u9042\u3052\u305f\u3053\u3068\u3092\u8a73\u8ff0\u3057\u3066\u3044\u307e\u3059\u3002\u5177\u4f53\u7684\u306b\u306f\u3001\u30ea\u30fc\u30de\u30f3\u30bc\u30fc\u30bf\u95a2\u6570\u306e\u30bc\u30ed\u70b9\u306e\u5272\u5408\u306b\u95a2\u3059\u308b\u9577\u5e74\u306e\u4e0b\u9650\u3092\u3001\u3053\u308c\u307e\u3067\u306e41.6%\u304b\u308967.2%\u306b\u5411\u4e0a\u3055\u305b\u307e\u3057\u305f\u3002<\/p>\n<p>\u3053\u306e\u6210\u679c\u306f\u3001<strong>Claude<\/strong>\u304c<strong>\u30ea\u30fc\u30de\u30f3\u4e88\u60f3<\/strong>\u81ea\u4f53\u3092\u8a3c\u660e\u3059\u308b\u306b\u306f\u81f3\u3089\u306a\u304b\u3063\u305f\u3082\u306e\u306e\u3001\u305d\u306e\u8a66\u307f\u306e\u4e2d\u3067\u4e88\u671f\u305b\u305a\u9054\u6210\u3055\u308c\u305f\u3082\u306e\u3067\u3059\u3002<strong>Claude<\/strong>\u306f\u3001\u7d0460\u306e\u30b5\u30d6\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u3092\u8abf\u6574\u3057\u30012,400\u306e\u30b7\u30a7\u30eb\u30b3\u30de\u30f3\u30c9\u3068\u6570\u767e\u306ePython\u30b9\u30af\u30ea\u30d7\u30c8\u3092\u5b9f\u884c\u3059\u308b\u3068\u3044\u3046\u3001\u96c6\u4e2d\u7684\u306a\u8a08\u7b97\u30d7\u30ed\u30bb\u30b9\u3092\u901a\u3058\u3066\u3053\u306e\u767a\u898b\u306b\u81f3\u308a\u307e\u3057\u305f\u3002\u3053\u306e\u30d7\u30ed\u30bb\u30b9\u3067\u306f\u3001\u6570\u5343\u306e\u6570\u5024\u30c1\u30a7\u30c3\u30af\u304c\u5b9f\u884c\u3055\u308c\u3001\u30b5\u30d6\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u540c\u58eb\u304c\u4e92\u3044\u306e\u4f5c\u696d\u3092\u30ec\u30d3\u30e5\u30fc\u3057\u307e\u3057\u305f\u3002<\/p>\n<p><strong>Anthropic<\/strong>\u306e\u6570\u5b66\u8005\u305f\u3061\u306f<strong>Claude<\/strong>\u306e\u8ad6\u6587\u3092\u691c\u8a3c\u3057\u3001\u305d\u306e\u7d50\u679c\u304c\u65e2\u5b58\u306e\u6570\u5b66\u7814\u7a76\u3001\u7279\u306b<strong>Baluyot<\/strong>\u3001<strong>Goldston<\/strong>\u3001<strong>Suriajaya<\/strong>\u3001<strong>Turnage-Butterbaugh<\/strong>\u3001\u304a\u3088\u3073<strong>Bombieri<\/strong>\u306e\u696d\u7e3e\u3068\u3069\u306e\u3088\u3046\u306b\u95a2\u9023\u3057\u3066\u3044\u308b\u304b\u3092\u660e\u3089\u304b\u306b\u3057\u307e\u3057\u305f\u3002\u3053\u306e\u767a\u898b\u306f\u3001AI\u30e2\u30c7\u30eb\u306e\u6570\u5b66\u7684\u80fd\u529b\u306e\u9032\u6b69\u306e\u901f\u3055\u3092\u793a\u3059\u6700\u65b0\u306e\u4f8b\u3067\u3042\u308a\u3001\u6570\u5b66\u8005\u306e\u30a2\u30a4\u30c7\u30a2\u306e\u5230\u9054\u7bc4\u56f2\u3092\u5e83\u3052\u308b\u53ef\u80fd\u6027\u3092\u79d8\u3081\u3066\u3044\u307e\u3059\u3002<\/p>\n<p><strong>Claude<\/strong>\u81ea\u8eab\u3082\u5f53\u521d\u306f\u305d\u306e\u767a\u898b\u306b\u61d0\u7591\u7684\u3067\u3057\u305f\u304c\u3001\u52b1\u307e\u3057\u306e\u30d7\u30ed\u30f3\u30d7\u30c8\u306b\u3088\u3063\u3066\u6700\u7d42\u7684\u306b\u7d50\u679c\u3092\u5c0e\u304d\u51fa\u3057\u307e\u3057\u305f\u3002\u3053\u308c\u306f\u3001AI\u30e2\u30c7\u30eb\u304c\u81ea\u8eab\u306e\u80fd\u529b\u3092\u904e\u5c0f\u8a55\u4fa1\u3059\u308b\u53ef\u80fd\u6027\u3092\u793a\u5506\u3057\u3066\u304a\u308a\u3001AI\u306e\u9032\u6b69\u306e\u901f\u5ea6\u3092\u4eba\u9593\u304c\u904e\u5c0f\u8a55\u4fa1\u3057\u3066\u3044\u308b\u53ef\u80fd\u6027\u3082\u793a\u5506\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n<p>\ud83d\udc49 <strong><a href=\"https:\/\/www.anthropic.com\/research\/riemann-zeta\" target=\"_blank\" rel=\"noopener\">Anthropic \u3067\u8a18\u4e8b\u5168\u6587\u3092\u8aad\u3080<\/a><\/strong><\/p>\n<\/div>\n<ul>\n<li><strong>\u8981\u70b9:<\/strong> Anthropic's Claude significantly advanced the lower bound for the Riemann zeta function's zeros satisfying the Riemann hypothesis, showcasing rapid progress in AI's mathematical capabilities through extensive multi-agent computation.<\/li>\n<li><strong>\u8457\u8005:<\/strong> Editorial Staff<\/li>\n<\/ul>\n<blockquote class=\"wp-block-quote\"><p><span>English Summary:<\/span><\/p>\n<p>A research article from <strong>Anthropic<\/strong> details a groundbreaking advancement made by an unreleased research version of <strong>Claude<\/strong> concerning one of mathematics' most famous unsolved problems, the <strong>Riemann hypothesis<\/strong>. Specifically, <strong>Claude<\/strong> improved a longstanding lower bound for the fraction of zeros of the Riemann zeta function that satisfy the Riemann hypothesis, increasing it from 41.6% to 67.2%.<\/p>\n<p>This achievement, though not a proof of the <strong>Riemann hypothesis<\/strong> itself, emerged unexpectedly during <strong>Claude's<\/strong> attempt to tackle the problem. <strong>Claude<\/strong> arrived at this finding through an intensive computational process, coordinating approximately 60 subagents, executing 2,400 shell commands, and writing hundreds of Python scripts. This process involved thousands of numerical checks and peer-review among the subagents.<\/p>\n<p>Mathematicians at <strong>Anthropic<\/strong> validated <strong>Claude's<\/strong> paper, clarifying how its results integrate with prior mathematical research, particularly the works of <strong>Baluyot<\/strong>, <strong>Goldston<\/strong>, <strong>Suriajaya<\/strong>, <strong>Turnage-Butterbaugh<\/strong>, and <strong>Bombieri<\/strong>. This discovery serves as the latest example of the rapid progress in AI models' mathematical capabilities, demonstrating their potential to extend the reach of mathematicians' ideas.<\/p>\n<p>Initially, <strong>Claude<\/strong> itself was skeptical of its own finding, but through encouraging prompts, it ultimately produced the result. This suggests that AI models might underestimate their own capabilities, potentially mirroring how humans might underestimate the pace of AI progress.<\/p>\n<\/blockquote>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>OpenAI\u306eGPT-5.6-Cyber\u306b\u3088\u308b\u30b5\u30a4\u30d0\u30fc\u9632\u5fa1\u5f37\u5316\u3001Anthropic\u306eClaude\u306b\u3088\u308b\u30ea\u30fc\u30de\u30f3\u4e88\u60f3\u95a2\u9023\u306e\u6570\u5b66\u7684\u9032\u5c55\u3001\u305d\u3057\u3066AI\u30e2\u30c7\u30eb\u306e\u5185\u90e8\u601d\u8003\u3092\u89e3\u660e\u3059\u308b\u65b0\u6280\u8853\u306b\u3064\u3044\u3066\u6df1\u6398\u308a\u3002<\/p>\n","protected":false},"author":1,"featured_media":6,"comment_status":"","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"sns_share_botton_hide":"","vkExUnit_sns_title":"","_vk_print_noindex":"","sitemap_hide":"","_veu_custom_css":"","veu_display_promotion_alert":"","vkexunit_cta_each_option":"","_lightning_design_setting":[],"footnotes":""},"categories":[3],"tags":[8,17,58,16,57,56,26,274,15],"class_list":["post-1136","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-columns","tag-ai","tag-anthropic","tag-llm","tag-openai","tag-56","tag-26","tag-274"],"veu_head_title_object":{"title":"","add_site_title":""},"_links":{"self":[{"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/posts\/1136","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/comments?post=1136"}],"version-history":[{"count":0,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/posts\/1136\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/media\/6"}],"wp:attachment":[{"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/media?parent=1136"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/categories?post=1136"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/itexplore.org\/jp\/wp-json\/wp\/v2\/tags?post=1136"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}