DeepSeek\u3084OpenAI\u306e\u3088\u3046\u306a\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(large language model; LLM)\u3092\u4f7f\u3063\u305f\u751f\u6210AI\u306f\u3001\u5c11\u306a\u304f\u3068\u3082\u30b9\u30de\u30db\u306e\u767b\u5834\u3068\u540c\u3058\u304f\u3089\u3044\u306e\u30a4\u30f3\u30d1\u30af\u30c8\u3092\u4e0e\u3048\u308b\u3068\u4e88\u60f3\u3057\u3066\u3044\u307e\u3059\u3002\u79c1\u306e\u3088\u3046\u306a\u65e5\u66dc\u30d7\u30ed\u30b0\u30e9\u30de\u3068\u3057\u3066\u3082\u9762\u767d\u305d\u3046\u306a\u984c\u6750\u304c\u3044\u304f\u3089\u3067\u3082\u3042\u308a\u307e\u3059\u3002
\nDeepSeek\u306e\u8a71\u984c\u306e\u8ad6\u6587\u306f\u4e0b\u8a18\u3067\u3059\u3002
\nhttps:\/\/github.com\/deepseek-ai\/DeepSeek-R1\/blob\/main\/DeepSeek_R1.pdf
\n\u308f\u304b\u308a\u3084\u3059\u3044\u89e3\u8aac\u304c\u4e0b\u8a18\u3067\u3059\u3002
\nhttps:\/\/zenn.dev\/asap\/articles\/34237ad87f8511
\n\u79c1\u306f\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u5b66\u8005\u3068\u4e00\u7dd2\u306bAI\u306b\u6709\u6a5f\u5316\u5b66\u306e\u554f\u984c\u3092\u89e3\u304b\u305b\u308b\u672c\u3092\u66f8\u3053\u3046\u3068\u3057\u3066\u3044\u307e\u3059\u304c\u3001DeepSeek\u306e\u5b66\u7fd2\u904e\u7a0b\u306b\u306f\u304b\u306a\u308a\u30d2\u30f3\u30c8\u304c\u3042\u308a\u305d\u3046\u3067\u3059\u3002\u8a08\u7b97\u8cc7\u6e90\u304c\u554f\u984c\u306b\u306a\u308a\u305d\u3046\u3067\u3001\u3069\u3053\u304b\u3089\u8abf\u9054\u3059\u308b\u304b\u982d\u304c\u75db\u3044\u3067\u3059\u3002
\nDeepSeek\u306f1\u500b500\u4e07\u5186\u306eNVIDIA-H800\u30922000\u53f0\u4f7f\u3063\u30662\u304b\u6708\u3067\u4f5c\u3089\u308c\u305f\u3068\u306e\u3053\u3068\u3067\u3059\u3002\u84b8\u7559\u30e2\u30c7\u30eb\u306f\u500b\u4eba\u306e\u30d1\u30bd\u30b3\u30f3\u30ec\u30d9\u30eb\u3067\u52d5\u304f\u305d\u3046\u3067\u3059\u3002\u30de\u30a4\u30cb\u30f3\u30b0\u7528\u306e7\u5e74\u843d\u3061\u306eCPU\uff0828\u30b3\u30a2\uff09\u304cAlibaba\u3067\u683c\u5b89(50\u4e07\u5186\u21922000\u5186)\u3067\u58f2\u3089\u308c\u3066\u3044\u3066\u3001\u8cb7\u3063\u3066\u307f\u307e\u3057\u305f\u304c\u6700\u65b0\u578b(>10\u4e07\u5186)\u306b\u6bd4\u3079\u3066\u901f\u304f\u3042\u308a\u307e\u305b\u3093\u3002\u30b2\u30fc\u30df\u30f3\u30b0\u7528\u306eGPU\u3068\u7d44\u307f\u5408\u308f\u305b\u3066\u3001\u5316\u5b66\u306e\u4eba\u306bAI\u3092\u3001AI\u306e\u4eba\u306b\u5316\u5b66\u3092\u3001\u305d\u308c\u305e\u308c\u306e\u96f0\u56f2\u6c17\u3092\u4f1d\u3048\u3089\u308c\u308b\u304f\u3089\u3044\u306e\u672c\u304c\u66f8\u3051\u306a\u3044\u304b\u69cb\u60f3\u3092\u7df4\u3063\u3066\u3044\u307e\u3057\u305f\u304c\u3001DeepSeek\u767b\u5834\u3067\u8a08\u753b\u5909\u66f4\u3001\u305c\u3072LLM\u3092\u53d6\u308a\u5165\u308c\u3066\u307f\u305f\u3044\u3067\u3059\u3002<\/p>\n
\u82f1\u8a9e\u306f\u3000https:\/\/github.com\/deepseek-ai\/DeepSeek-R1\/blob\/main\/DeepSeek_R1.pdf
\na model trained via large-scale reinforcement learning without supervised fine-tuning\u3000\u6559\u5e2b\u30c7\u30fc\u30bf\u306b\u3088\u308b\u5fae\u8abf\u6574\u306a\u3057\u306e\u5927\u898f\u6a21\u306a\u5f37\u5316\u5b66\u7fd2
\nas a preliminary step\u3000\u4e88\u5099\u7684\u306a\u6bb5\u968e
\n\u201don AIME2024, …, On MATH-500, it attains an impressive score of 97.3%, performing on par with OpenAI-o1-1217 and significantly outperforming other models.\u201d
\nAIME2024 AI\u306e\u5b66\u4f1a\u3067\u51fa\u3055\u308c\u305f\u6570\u5b66\u306e\u8ab2\u984c\u306e\u3088\u3046\u3067\u3059\u3002https:\/\/mathlog.info\/articles\/fA5Hi3FJiWbRjfRFh8qz\u3000\u306b\u548c\u8a33\u304c\u3042\u308a\u307e\u3059\u3002\u7d50\u69cb\u96e3\u3057\u3044\u3067\u3059\u3002\u4f8b\uff1az\u304c |z|=4\u3092\u6e80\u305f\u3059\u8907\u7d20\u6570\u3067\u3042\u308b\u3068\u304d\u3001(75+117i)z+(96+144i)\/z\u306e\u6700\u5927\u5024\u3092\u6c42\u3081\u3066\u304f\u3060\u3055\u3044\u3001\u306a\u3069\u3002\u56f3\u5f62\u554f\u984c\u3082\u3042\u308a\u307e\u3059\u3002\u672c\u5bb6\u306f\u3000https:\/\/artofproblemsolving.com\/community\/c3370201_2024_aime\u3000\u3000\u3000\u3053\u308c\u3092DeepSeek\u3084OpenAI\u306f\u89e3\u3051\u308b\u306e\u3067\u3059\u306d\u3002DeepSeek\u306e\u6b63\u7b54\u738786.7\uff05\u3060\u305d\u3046\u3067\u3059\u3002
\nMATH-500\u3000https:\/\/huggingface.co\/datasets\/HuggingFaceH4\/MATH-500\u3000\u3067\u3001500\u554f\u3042\u308a\u307e\u3059\u3002DeepSeek\u304c\u6b63\u7b54\u738795.9%\u3068\u3044\u3046\u306e\u306f\u6050\u308c\u5165\u308a\u307e\u3059\u3002\u7686\u3055\u3093\u306f\u3069\u306e\u304f\u3089\u3044\u3067\u304d\u307e\u3059\u304b\uff1f
\n“aha moment”\u306f\u3001\u221a(a-\u221a(a+x))=x\u306e\u5b9f\u6570\u89e3\u306e\u7dcf\u548c\u3092\u6c42\u3081\u3088\u3001\u3068\u3044\u3046\u554f\u984c\u3092\u89e3\u3044\u3066\u3044\u308b\u3068\u304d\u306b\u3001”Wait, wait. Wait. That’s an aha moment I can flag here.”\u3068AI\u304c\u51fa\u529b\u3057\u3066\u8003\u3048\u76f4\u3057\u59cb\u3081\u305f\u305d\u3046\u3067\u3059\u3002
\n\u3053\u308c\u306f\u30bb\u30ea\u30d5\uff08meme \u30df\u30fc\u30e0\u3001\u96c6\u56e3\u5185\u306e\u6a21\u5023\u884c\u52d5\u3001\u30a4\u30f3\u30bf\u30fc\u30cd\u30c3\u30c8\u4e0a\u3067\u62e1\u6563\u3055\u308c\u308b\u30b3\u30f3\u30c6\u30f3\u30c4\uff09\u3068\u3057\u3066\u3057\u3070\u3089\u304f\u6d41\u884c\u308b\u304b\u3082\u3057\u308c\u307e\u305b\u3093\u3002\u601d\u8003\u7d4c\u904e\u3092\u6587\u7ae0\u3067\u51fa\u529b\u3059\u308b\u306e\u306f\u6709\u52b9\u3067\u3057\u3087\u3046\u306d\u3002<\/p>\n
\u203b\u3053\u306e\u8ad6\u6587\u306b\u306f\u3053\u308c\u3089\u554f\u984c\u3092\u89e3\u3051\u308b\u3088\u3046\u306b\u5b66\u7fd2\u3055\u305b\u305f\u65b9\u6cd5\u304c\u66f8\u3044\u3066\u3042\u308b\u3088\u3046\u306a\u306e\u3067\u3001\u7cbe\u8aad\uff08\u6709\u5fd7\u3067\u8f2a\u8b1b\uff1f\uff09\u3057\u305f\u3044\u3067\u3059\u3002\u767a\u8868\u304b\u30892\u9031\u9593\u4ee5\u4e0a\u305f\u3063\u3066\u3044\u308b\u306e\u3067\u3001\u4e16\u754c\u4e2d\u3067\u3044\u308d\u3044\u308d\u3084\u3063\u3066\u3044\u308b\u3053\u3068\u3067\u3057\u3087\u3046\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
DeepSeek\u3084OpenAI\u306e\u3088\u3046\u306a\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(large language model; LLM)\u3092\u4f7f\u3063\u305f\u751f\u6210AI\u306f\u3001\u5c11\u306a\u304f\u3068\u3082\u30b9\u30de\u30db\u306e\u767b\u5834\u3068\u540c\u3058\u304f\u3089\u3044\u306e\u30a4\u30f3\u30d1\u30af\u30c8\u3092\u4e0e\u3048\u308b\u3068\u4e88\u60f3\u3057\u3066\u3044\u307e\u3059\u3002\u79c1\u306e\u3088\u3046\u306a\u65e5\u66dc\u30d7\u30ed\u30b0\u30e9\u30de\u3068\u3057\u3066\u3082\u9762\u767d\u305d\u3046\u306a\u984c\u6750\u304c\u3044\u304f\u3089\u3067\u3082\u3042\u308a\u307e\u3059\u3002 DeepSeek\u306e\u8a71\u984c\u306e\u8ad6\u6587\u306f\u4e0b\u8a18\u3067\u3059\u3002 https:\/\/github.com\/deepseek-ai\/DeepSeek-R1\/blob\/main\/DeepSeek_R1.pdf \u308f\u304b\u308a\u3084\u3059\u3044\u89e3\u8aac\u304c\u4e0b\u8a18\u3067\u3059\u3002 https:\/\/zenn.dev\/asap\/articles\/34237ad87f8511 \u79c1\u306f\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u5b66\u8005\u3068\u4e00\u7dd2\u306b…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"footnotes":""},"categories":[60,42],"tags":[23,5],"class_list":["post-2134","post","type-post","status-publish","format-standard","hentry","category-computer","category-tech","tag-computer","tag-tech"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=\/wp\/v2\/posts\/2134","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=2134"}],"version-history":[{"count":0,"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=\/wp\/v2\/posts\/2134\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=2134"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=2134"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=2134"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}