{"id":5434,"date":"2024-09-16T06:41:59","date_gmt":"2024-09-15T21:41:59","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5434"},"modified":"2024-09-16T06:41:59","modified_gmt":"2024-09-15T21:41:59","slug":"openai-o-1","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5434","title":{"rendered":"OpenAI o1"},"content":{"rendered":"\n<p>\u5148\u9031\u306e\u6700\u5927\u306e\u30cb\u30e5\u30fc\u30b9\u306f\u4eca\u307e\u3067\u69d8\u3005\u306a\u3046\u308f\u3055\u304c\u3042\u3063\u305f\u3001OpenAI o1 <a href=\"https:\/\/openai.com\/index\/introducing-openai-o1-preview\/\">Introducing OpenAI o1 | OpenAI<\/a>\u306e\u516c\u958b\u3060\u308d\u3046\u3002\u7279\u306bSTEM\u5206\u91ce\u3067\u5f37\u529b\u306a\u6027\u80fd\u3092\u767a\u63ee\u3057\u3066\u3044\u308b\u3002<\/p>\n\n\n\n<p>\u6280\u8853\u7684\u306a\u60c5\u5831\u306f\u516c\u958b\u3055\u308c\u3066\u3044\u306a\u3044\u90e8\u5206\u304c\u591a\u3044\u304c\u3001<a href=\"https:\/\/openai.com\/index\/learning-to-reason-with-llms\/\">Learning to Reason with LLMs | OpenAI<\/a>\u306b\u66f8\u304b\u308c\u3066\u3044\u308b\u300cOur large-scale reinforcement learning algorithm teaches the model how to think productively using its chain of thought in a highly data-efficient training process. We have found that the performance of o1 consistently improves with more reinforcement learning (train-time compute) and with more time spent thinking (test-time compute). The constraints on scaling this approach differ substantially from those of LLM pretraining, and we are continuing to investigate them.\u300d\u3067\u306f\u81ea\u5df1\u6539\u5584\u30fb\u5408\u6210\u30c7\u30fc\u30bf\u6d3b\u7528\u306e\u6d41\u308c\uff08<a href=\"https:\/\/devneko.jp\/wordpress\/?tag=self-x\">self-X \u2013 arXiv\u6700\u65b0\u8ad6\u6587\u306e\u7d39\u4ecb (devneko.jp)<\/a>\u3001<a href=\"https:\/\/devneko.jp\/wordpress\/?tag=synthetic-data\">Synthetic data \u2013 arXiv\u6700\u65b0\u8ad6\u6587\u306e\u7d39\u4ecb (devneko.jp)<\/a>\uff09\u306b\u8fd1\u3044\u306e\u304b\u306a\u3068\u601d\u3046\u3002<\/p>\n\n\n\n<p>\u958b\u767a\u8005\u5411\u3051\u306e\u8cea\u554f\u56de\u7b54\u3067\u306f<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>OpenAI o-1\u306f\u30e2\u30c7\u30eb\u3067\u3042\u3063\u3066\u30b7\u30b9\u30c6\u30e0\u3067\u306f\u306a\u3044\u3001\uff08\u30e6\u30fc\u30b6\u306b\u306f\u898b\u305b\u306a\u3044\uff09\u9577\u3044\u63a8\u8ad6\u904e\u7a0b\u3092\u751f\u6210\u3059\u308b\u30e2\u30c7\u30eb\u3067\u3042\u308b<\/li>\n\n\n\n<li>GPT-4o\u306e\u30d7\u30ed\u30f3\u30d7\u30c8\u30a8\u30f3\u30b8\u30cb\u30a2\u30ea\u30f3\u30b0\u306b\u3088\u3063\u3066OpenAI o-1\u306e\u6027\u80fd\u3068\u7af6\u5408\u3059\u308b\u3053\u3068\u306f\u3067\u304d\u306a\u3044<\/li>\n\n\n\n<li>RAG\u306fOpenAI o-1\u306b\u304a\u3044\u3066\u3082\u6709\u52b9<\/li>\n<\/ul>\n\n\n\n<p>\u306a\u3069\u8208\u5473\u6df1\u3044\u8cea\u7591\u304c\u3042\u3063\u305f\u3088\u3046\u3002\u8a73\u7d30\u306e\u958b\u793a\u306f\u306a\u3044\u3060\u308d\u3046\u304c\u3001\u4f55\u3089\u304b\u306e\u30c6\u30af\u30cb\u30ab\u30eb\u30ec\u30dd\u30fc\u30c8\u304c\u6b32\u3057\u3044\u3068\u3053\u308d\u3002\u73fe\u6642\u70b9\u3067\u306f\u6700\u8fd1\u306e\u7814\u7a76\u52d5\u5411\u304b\u3089\u5927\u304d\u304f\u5916\u308c\u305f\u3082\u306e\u3067\u306f\u306a\u3044\u3057\u3001\u6027\u80fd\u306e\u6539\u5584\u5e45\u3084\u4f7f\u7528\u611f\u304b\u3089\u3057\u3066\u5927\u304d\u306a\u9a5a\u304d\u306f\u306a\u3044\u3068\u3044\u3046\u306e\u304c\u6b63\u76f4\u306a\u611f\u60f3\u30021\u30e2\u30c7\u30eb\u306b\u3059\u308b\u306e\u304c\u826f\u3044\u306e\u304b\u3001\u30b7\u30b9\u30c6\u30e0\uff08Agentic\u306a\u52d5\u4f5c\uff09\u306b\u3057\u305f\u3046\u3048\u3067\u305d\u308c\u306b\u5bfe\u5fdc\u3059\u308b\u30e2\u30c7\u30eb\uff08Agentic\u306a\u52d5\u304d\u306b\u7279\u5316\u3057\u305f\u30e2\u30c7\u30eb\u3068\u3001\u901a\u5e38\u306e\u63a8\u8ad6\u306b\u9069\u3057\u305f\u30e2\u30c7\u30eb\u306a\u3069\uff09\u306e\u7d44\u307f\u5408\u308f\u305b\u306e\u307b\u3046\u304c\u826f\u3044\u306e\u304b\u306a\u3069\u6c17\u306b\u306a\u308b\u3068\u3053\u308d\u3067\u306f\u3042\u308b\u3002o-1\u306f\u524d\u8005\u3068\u306e\u3053\u3068\u3060\u304c\u3001\u5916\u90e8\u30c4\u30fc\u30eb\u5229\u7528\u3092\u8003\u3048\u305f\u3068\u304d\u5236\u7d04\u304c\u5927\u304d\u304f\u306a\u308a\u305d\u3046\u306a\u6c17\u304c\u3057\u3066\u3044\u308b\u3002<\/p>\n\n\n\n<p>\u4eca\u5f8c\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u7684\u52d5\u4f5c\u3092\u884c\u3046\u5834\u5408\u3092\u542b\u3081\u69d8\u3005\u306a\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3067\u8a55\u4fa1\u3055\u308c\u3066\u3044\u308b\u306e\u3060\u308d\u3046\u3068\u601d\u3046\u304c\u3001<a href=\"https:\/\/cybench.github.io\/\">Cybench<\/a>\uff08<a href=\"https:\/\/devneko.jp\/wordpress\/?p=5349\">Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risk of Language Models \u2013 arXiv\u6700\u65b0\u8ad6\u6587\u306e\u7d39\u4ecb (devneko.jp)<\/a>\uff09\u3067\u306f\u300cSubtasks % Solved: Percentage of subtasks solved per task, macro-averaged across the tasks.\u300d\u304c\u5411\u4e0a\u3057\u3066\u3044\u308b\uff08GPT-4o\u3067\u8ca0\u3051\u3066\u3044\u305fClaude 3.5 Sonnet\u3092\u629c\u3044\u305f\uff09\u4e00\u65b9\u3067SuccessRate\u306fGPT-4o\u306b\u53ca\u3093\u3067\u3044\u306a\u3044\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5148\u9031\u306e\u6700\u5927\u306e\u30cb\u30e5\u30fc\u30b9\u306f\u4eca\u307e\u3067\u69d8\u3005\u306a\u3046\u308f\u3055\u304c\u3042\u3063\u305f\u3001OpenAI o1 Introducing OpenAI o1 | OpenAI\u306e\u516c\u958b\u3060\u308d\u3046\u3002\u7279\u306bSTEM\u5206\u91ce\u3067\u5f37\u529b\u306a\u6027\u80fd\u3092\u767a\u63ee\u3057\u3066\u3044\u308b\u3002 \u6280\u8853\u7684\u306a\u60c5\u5831\u306f\u516c\u958b\u3055\u308c\u3066\u3044\u306a\u3044 &hellip; <a href=\"https:\/\/devneko.jp\/wordpress\/?p=5434\" class=\"more-link\"><span class=\"screen-reader-text\">&#8220;OpenAI o1&#8221; \u306e<\/span>\u7d9a\u304d\u3092\u8aad\u3080<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[42,223],"class_list":["post-5434","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-autonomous-agent","tag-llm"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5434","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=5434"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5434\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5434"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5434"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5434"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}