\u4f3c\u305f\u3088\u3046\u306a\u8a71\u304c\u7d9a\u3044\u3066\u6050\u7e2e\u3067\u3059\u304c\u3001\u4eca\u9031\u306f\u5148\u9031\u5404\u65b9\u9762\u306b\u6fc0\u9707\u304c\u8d70\u3063\u305fDeepseek\u554f\u984c\u3092\u3068\u308a\u3042\u3052\u307e\u3059\u3002Deepseek\u306f\u4e2d\u56fd\u306e\u4eba\u5de5\u77e5\u80fd\u7814\u7a76\u6240\u300c\u676d\u5dde\u6df1\u5ea6\u6c42\u7d22\u4eba\u5de5\u667a\u80fd\u57fa\u7840\u6280\u672f\u7814\u7a76\u6709\u9650\u516c\u53f8\u300d\u306e\u82f1\u8a9e\u540d\u3067\u3059\u3002\u305d\u3053\u304c\u5148\u3005\u9031\u306e1\u670820\u65e5\u306b\u767a\u8868\u3057\u305fDeepSeek-R1\u304c\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u306e\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30ebAI(large language model; LLM)\u3067\u3001\u901a\u5e38\u306e\u8a00\u8a9e\u30d7\u30ed\u30f3\u30d7\u30c8\u306b\u3088\u308b\u65e2\u5b58\u306e\u5546\u7528LLM\u3068\u540c\u6027\u80fd\u306e\u5fdc\u7b54\u306e\u307b\u304b\u306b\u6570\u5b66\u7684\u63a8\u8ad6\u306b\u9577\u3051\u3066\u3044\u308b\u3068\u3044\u3046\u3053\u3068\u3067\u5927\u9a12\u304e\u306b\u306a\u308a\u307e\u3057\u305f\u3002\u540c\u793e\u306e\u793e\u54e1\u306f\u4e0b\u8a18\u306b\u3088\u308b\u3068200\u4eba\u3060\u305d\u3046\u3067\u3059\u3002
\nhttps:\/\/ja.wikipedia.org\/wiki\/DeepSeek
\nhttps:\/\/jbpress.ismedia.jp\/articles\/-\/86390
\nLLM\u306fOpenAI\u793e\u3001Google, Meta, Softbank\u95a2\u9023, NTT\u95a2\u9023\u7b49\u591a\u304f\u306e\u4f1a\u793e\u304c\u624b\u639b\u3051\u3066\u3044\u307e\u3059\u304c\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u3067\u306f\u306a\u304f\u305d\u308c\u306a\u308a\u306e\u4f7f\u7528\u6599\u3092\u3068\u308d\u3046\u3068\u3057\u3066\u3044\u307e\u3059\u3002\u305d\u306e\u53ce\u76ca\u6027\u3092\u671f\u5f85\u3057\u3066\u5927\u898f\u6a21\u6295\u8cc7\u304c\u884c\u308f\u308c\u308b\u3068\u5171\u306b\u7c73\u56fd\u306ebig tech\u3084\u30cf\u30fc\u30c9\u30a6\u30a7\u30a2\u306eNVIDIA\u7b49\u306eAI\u95a2\u9023\u682a\u304c\u5927\u304d\u304f\u4e0a\u6607\u3057\u3066\u304d\u307e\u3057\u305f\u304c\u3001DeepSeek\u304c\u683c\u5b89\uff08\u958b\u767a\u8cbb1\/20\uff09\u3067\u4f7f\u3048\u308b\u53ef\u80fd\u6027\u304c\u51fa\u3066\u304d\u305f\u306e\u3067\u682a\u4fa1\u306e\u4e0b\u843d\u3084\u653f\u6cbb\u7684\u5bfe\u5fdc\u306a\u3069\u69d8\u3005\u306a\u60c5\u5831\u304c\u3068\u3073\u304b\u3063\u3066\u3044\u307e\u3059\u3002\u79c1\u306f\u6570\u4eba\u3067AI\u306e\u5316\u5b66\u3078\u306e\u5fdc\u7528\u306e\u672c\u3092\u66f8\u3053\u3046\u3068\u672c\u8170\u3092\u5165\u308c\u59cb\u3081\u305f\u3068\u3053\u308d\u3067\u3059\u304c\u3001LLM\u95a2\u9023\u306e\u9032\u5c55\u304c\u901f\u304f\u60c5\u5831\u306b\u8ffd\u3044\u3064\u304f\u306e\u306b\u82e6\u52b4\u3057\u307e\u3059\u3002DeepSeek\u3067\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u304c\u51fa\u3066\u304d\u305f\u306e\u3067\u30bd\u30fc\u30b9\u30b3\u30fc\u30c9\u3092\u52c9\u5f37\u3057\u3066\u5b9f\u969b\u306b\u4f7f\u3063\u3066\u307f\u305f\u3044\u3068\u601d\u3044\u307e\u3059\u3002CUDA\u3088\u308a\u4f4e\u30ec\u30d9\u30eb\u306e\u8a00\u8a9e\u3092\u4f7f\u3063\u3066\u8f38\u51fa\u5236\u9650\u3055\u308c\u306a\u3044\u9045\u3044\u534a\u5c0e\u4f53\u3092\u4f7f\u3044\u3053\u306a\u3057\u3066\u3044\u308b\u3068\u3044\u3046\u3046\u308f\u3055\u3082\u3042\u308b\u306e\u3067\u8208\u5473\u6df1\u3044\u3067\u3059\u3002<\/p>\n
\u82f1\u8a9e\u306f\u3000https:\/\/www.reddit.com\/r\/singularity\/comments\/1icwl73\/notes_on_deepseek_r1_just_how_good_it_is_compared\/?rdt=55391\u3000\u304b\u3089\u3002
\n“Finally, there is a model worthy of the hype it has been getting.”\u3000\u3064\u3044\u306b\u3001\u3046\u3051\u3066\u3044\u308b\u904e\u5270\u306a\u8a55\u5224\u306b\u5024\u3059\u308b\u30e2\u30c7\u30eb\u304c\u51fa\u305f\u3002
\nhype \u30cf\u30a4\u30d7\u3000 \u8a87\u5927\u5e83\u544a\u3001\u904e\u5270\u5ba3\u4f1d
\n“the general public seems excited about this, while the big AI labs are probably scrambling. It feels like things are about to speed up in the AI world. And it’s all thanks to this new DeepSeek-R1 model and how they trained it. ”
\nthe general public \u5927\u8846
\nscrambling \u6df7\u4e71\u3057\u3066\u3044\u308b\u3000> scramble 1.\u9019\u3044\u4e0a\u304c\u308b 2.\u7dca\u6025\u767a\u9032\uff08\u30b9\u30af\u30e9\u30f3\u30d6\u30eb\uff09\u3059\u308b 3.\u76d7\u8074\u3055\u308c\u306a\u3044\u3088\u3046\u306b\u6697\u53f7\u5316\u3059\u308b\u30014.\u4e71\u96d1\u306b\u96c6\u3081\u308b\u30015.\u6df7\u4e71\u3055\u305b\u308b\u30016.\u30b9\u30af\u30e9\u30f3\u30d6\u30eb\u30a8\u30c3\u30b0\u306b\u3059\u308b
\n\u201dThe model uses \u201cAha moments\u201d as pivot tokens to reflect and reevaluate answers during CoT.\u201d
\nAha moments \u306f\u300c\u30a2\u30cf\u4f53\u9a13\u300d\u3067\u4e00\u6642\u671f\u6709\u540d\u306b\u306a\u308a\u307e\u3057\u305f\u3002\u3053\u308c\u3092\u4f7f\u3063\u3066\u3044\u308b\u3068\u3059\u308b\u3068\u3001AI\u306e\u81ea\u6211\u306e\u76ee\u899a\u3081\u306b\u3042\u3068\u4e00\u6b69\u306e\u3088\u3046\u306a\u6c17\u304c\u3057\u3066\u3057\u307e\u3044\u307e\u3059\u3002
\npivot \u652f\u70b9
\ntoken \u30c8\u30fc\u30af\u30f3\u3000LLM\u306e\u5206\u91ce\u3067\u306f\u610f\u5473\u5358\u4f4d\u3068\u3044\u3046\u610f\u5473\u3067\u4f7f\u3063\u3066\u3044\u308b\u3088\u3046\u3067\u3059\u3002
\nCoT \uff1d Chain-of-Thought \u3060\u305d\u3046\u3067\u3059\u3002AI\u5206\u91ce\u3067\u3044\u308f\u3086\u308b\u30d7\u30ed\u30f3\u30d7\u30c8\u30a8\u30f3\u30b8\u30cb\u30a2\u30ea\u30f3\u30b0\u3067\u554f\u984c\u3092\u5c0f\u3055\u3044\u8ad6\u7406\u7684\u30b9\u30c6\u30c3\u30d7\u306b\u5206\u89e3\u3059\u308b\u3053\u3068\u3002
\n\u201dSo, for this, I tested (DeepSeek\u306e) r1 and \uff08OpenAI\u306e GPT4\u306e\uff09o1 side by side on complex reasoning, math, coding, and creative writing problems. These are the questions that o1 solved only or by none before. It is better than o1-preview but a notch below o1.\u201d
\ncomplex reasoning \u8907\u96d1\u306a\u63a8\u8ad6
\ncreative writing problem \u5275\u9020\u7684\u4f5c\u6587\u554f\u984c
\na notch below o1 \u76ee\u76db\u308a\u4e00\u3064\u5206GPT4-o1\u3088\u308a\u6027\u80fd\u304c\u4f4e\u3044
\n“Writing: This is where R1 takes the lead. It gives the same vibes as early Opus. It\u2019s free, less censored, has much more personality, is easy to steer, and is very creative compared to the rest, even o1-pro.”
\nvibes = a person’s emotional state or the atmosphere of a place as communicated to and felt by others.
\nOpus \u4f5c\u66f2\u5bb6\u306e\u4f5c\u54c110\u756a=Opus 10 \u306e\u3088\u3046\u306b\u4f7f\u3046\u5358\u8a9e\u3067\u3059\u304c\u3001Opus.ai\u3068\u3044\u3046\u4f1a\u793e\u304c\u3042\u308a\u3001\u4f5c\u6587\u3084\u4f5c\u66f2\u306eAI\u30b5\u30fc\u30d3\u30b9\u3092\u63d0\u4f9b\u3057\u3066\u3044\u307e\u3059\u3002
\n“Distillation works, small models like Qwen and Llama trained over r1 generated data show significant improvements. ”
\ndistillation \u84b8\u7559\u3000\u3067\u3059\u304c\u3001AI\u5206\u91ce\u306b\u304a\u3051\u308b\u5185\u5bb9\u306f\u660e\u65e5\u4ee5\u964d\u89e3\u8aac\u3057\u307e\u3059\u3002
\nQwen\u306f\u4e2d\u56fd\u30a2\u30ea\u30d0\u30d0\u793e\u306eLLM-AI\u3002 Llama\u306fLLM\u306e\u30a4\u30f3\u30bf\u30fc\u30d5\u30a7\u30fc\u30b9\u3068\u3057\u3066\u6a19\u6e96\u306e\u30bd\u30d5\u30c8\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
\u4f3c\u305f\u3088\u3046\u306a\u8a71\u304c\u7d9a\u3044\u3066\u6050\u7e2e\u3067\u3059\u304c\u3001\u4eca\u9031\u306f\u5148\u9031\u5404\u65b9\u9762\u306b\u6fc0\u9707\u304c\u8d70\u3063\u305fDeepseek\u554f\u984c\u3092\u3068\u308a\u3042\u3052\u307e\u3059\u3002Deepseek\u306f\u4e2d\u56fd\u306e\u4eba\u5de5\u77e5\u80fd\u7814\u7a76\u6240\u300c\u676d\u5dde\u6df1\u5ea6\u6c42\u7d22\u4eba\u5de5\u667a\u80fd\u57fa\u7840\u6280\u672f\u7814\u7a76\u6709\u9650\u516c\u53f8\u300d\u306e\u82f1\u8a9e\u540d\u3067\u3059\u3002\u305d\u3053\u304c\u5148\u3005\u9031\u306e1\u670820\u65e5\u306b\u767a\u8868\u3057\u305fDeepSeek-R1\u304c\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u306e\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30ebAI(large language model; LLM)\u3067\u3001\u901a\u5e38\u306e\u8a00\u8a9e\u30d7\u30ed\u30f3\u30d7\u30c8\u306b\u3088\u308b\u65e2\u5b58\u306e\u5546\u7528LLM\u3068\u540c\u6027\u80fd\u306e\u5fdc\u7b54\u306e\u307b\u304b\u306b\u6570\u5b66\u7684\u63a8\u8ad6\u306b\u9577\u3051\u3066\u3044\u308b\u3068\u3044\u3046\u3053\u3068\u3067\u5927\u9a12\u304e\u306b\u306a\u308a\u307e\u3057\u305f\u3002\u540c\u793e\u306e\u793e\u54e1\u306f\u4e0b\u8a18\u306b\u3088\u308b\u3068200\u4eba\u3060\u305d\u3046\u3067\u3059\u3002 https:\/\/ja.wikipedia.org\/wiki\/DeepSeek https:…<\/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":[47],"tags":[10],"class_list":["post-2131","post","type-post","status-publish","format-standard","hentry","category-worldresearchinstitutes","tag-worldresearchinstitutes"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=\/wp\/v2\/posts\/2131","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=2131"}],"version-history":[{"count":0,"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=\/wp\/v2\/posts\/2131\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=2131"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=2131"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.sekaiken.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=2131"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}