{"id":6482,"date":"2025-03-27T04:09:00","date_gmt":"2025-03-26T19:09:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6482"},"modified":"2025-03-27T04:09:00","modified_gmt":"2025-03-26T19:09:00","slug":"measuring-ai-ability-to-complete-long-tasks","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6482","title":{"rendered":"Measuring AI Ability to Complete Long Tasks\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Measuring AI Ability to Complete Long Tasks\u00a0<\/strong>[6.0]<br>\u4eba\u9593\u304c\u901a\u5e38\u3001AI\u30e2\u30c7\u30eb\u304c\u9054\u6210\u3067\u304d\u308b\u30bf\u30b9\u30af\u3092\u5b8c\u4e86\u3059\u308b\u306e\u306b\u8981\u3059\u308b\u6642\u9593\u309250%\u306e\u6210\u529f\u7387\u3067\u6e2c\u5b9a\u3057\u307e\u3059\u3002 Claude 3.7 Sonnet\u306e\u3088\u3046\u306a\u73fe\u5728\u306e\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2AI\u30e2\u30c7\u30eb\u306f50\u5206\u7a0b\u5ea6\u306750%\u30bf\u30a4\u30e0\u306e\u5730\u5e73\u7dda\u3092\u6301\u3064\u3002 AI\u30e2\u30c7\u30eb\u306e\u6642\u9593\u7684\u5730\u5e73\u7dda\u306e\u5897\u52a0\u306f\u3001\u3088\u308a\u4fe1\u983c\u6027\u304c\u9ad8\u304f\u3001\u30df\u30b9\u306b\u9069\u5fdc\u3059\u308b\u80fd\u529b\u306b\u3088\u3063\u3066\u5f15\u304d\u8d77\u3053\u3055\u308c\u3066\u3044\u308b\u3088\u3046\u306b\u601d\u308f\u308c\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2503.14499v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2503.14499v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Tue, 18 Mar 2025 17:59:31 GMT)<\/li>\n\n\n\n<li>\u300cthe time humans typically take to complete tasks that AI models can complete with 50% success rate\u300d\u3092\u5b9a\u7fa9\u3068\u3059\u308b\u300c50%-task-completion time horizon\u300d\u3068\u3044\u3046\u30e1\u30c8\u30ea\u30af\u30b9\u306e\u63d0\u6848\u3068\u691c\u8a0e\u3002\u300cOn these tasks, current frontier AI models such as Claude 3.7 Sonnet have a 50% time horizon of around 50 minutes\u300d\u3001\u300cFurthermore, frontier AI time horizon has been doubling approximately every seven months since 2019, though the trend may have accelerated in 2024.\u300d\u3068\u306e\u3053\u3068\u3002<\/li>\n\n\n\n<li>\u3069\u306e\u304f\u3089\u3044\u306e\u898f\u6a21\u306e\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u3092\u81ea\u52d5\u751f\u6210\u3067\u304d\u308b\u306e\u304b\uff1f\u3068\u3044\u3046\u610f\u5473\u3067\u306f\u53c2\u8003\u306b\u306a\u308b\u6307\u6a19\u3060\u3068\u601d\u3046\u3002\u300cFinally, we attempt to extrapolate the trend on our tasks to one-month (167 hours) AI (Section 7.1), finding that if both the trend continues and observed performance trends generalize to real-world tasks, an 80% confidence interval for the release date of AI that can complete 1-month long software tasks spans from late 2028 to early 2031\u300d\u3092\u3069\u3046\u8a55\u4fa1\u3059\u308b\u304b\u306f\u60a9\u307e\u3057\u3044\u304c\u3001\u4eba\u304c\u4e00\u304b\u6708\u304b\u3051\u3066\u958b\u767a\u3059\u308b\u30ec\u30d9\u30eb\u306e\u30bd\u30d5\u30c8\u30a6\u30a7\u30a2\u304c\u81ea\u52d5\u751f\u6210\u3067\u304d\u308b\u3088\u3046\u306b\u306a\u308b\u304b\u3082\u3001\u3068\u3044\u3046\u306e\u306f\u305d\u3046\u304b\u3082\u3057\u308c\u306a\u3044\u3068\u3044\u3046\u611f\u899a\u3082\u3042\u308b\u3002<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[647],"class_list":["post-6482","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-647"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6482","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=6482"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6482\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6482"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6482"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6482"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}