{"id":6931,"date":"2025-06-16T06:05:00","date_gmt":"2025-06-15T21:05:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6931"},"modified":"2025-06-15T20:32:43","modified_gmt":"2025-06-15T11:32:43","slug":"self-adapting-language-models","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6931","title":{"rendered":"Self-Adapting Language Models\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Self-Adapting Language Models&nbsp;<\/strong>[44.5]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306f\u5f37\u529b\u3060\u304c\u9759\u7684\u3067\u3042\u308a\u3001\u65b0\u3057\u3044\u30bf\u30b9\u30af\u3084\u77e5\u8b58\u3001\u4f8b\u306b\u5bfe\u5fdc\u3057\u3066\u91cd\u307f\u3092\u9069\u5fdc\u3059\u308b\u30e1\u30ab\u30cb\u30ba\u30e0\u304c\u6b20\u5982\u3057\u3066\u3044\u308b\u3002 \u6211\u3005\u306f,\u81ea\u5df1\u9069\u5fdc\u578bLSM(Self-Adapting LLMs, SEAL)\u3092\u5c0e\u5165\u3059\u308b\u3002 \u77e5\u8b58\u306e\u5b9a\u5f0f\u5316\u3068\u6570\u30b7\u30e7\u30c3\u30c8\u306e\u4e00\u822c\u5316\u306e\u5b9f\u9a13\u306b\u3088\u308a\u3001SEAL\u306f\u81ea\u5df1\u6307\u5411\u9069\u5fdc\u304c\u53ef\u80fd\u306a\u8a00\u8a9e\u30e2\u30c7\u30eb\u306b\u5411\u3051\u305f\u6709\u671b\u306a\u30b9\u30c6\u30c3\u30d7\u3067\u3042\u308b\u3053\u3068\u304c\u793a\u3055\u308c\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2506.10943v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2506.10943v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Thu, 12 Jun 2025 17:48:13 GMT)<\/li>\n\n\n\n<li>\u300cWe propose Self-Adapting LLMs (SEAL), a framework that enables language models to improve themselves by generating their own synthetic data and optimization parameters (\u201cself-edits\u201d) in re- sponse to new data. The model is trained to produce these self-edits directly through token generation with the data provided in the model\u2019s context. Self-edit generation is learned via reinforcement learning (RL) where the model is rewarded for generating self-edits (SE) that, when applied, improve the model\u2019s performance at the target task.\u300d\u3068\u3044\u3046\u81ea\u5df1\u9069\u5408\u3001\u81ea\u5df1\u9032\u5316\u3001\u81ea\u5df1\u6539\u5584\u306e\u30a2\u30d7\u30ed\u30fc\u30c1\u3002SQuAD\u3084ARC-AGI benchmark\uff08\u306e\u30b5\u30d6\u30bb\u30c3\u30c8\uff09\u3092\u7528\u3044\u3066\u52b9\u679c\u3092\u691c\u8a3c\u3057\u3066\u3044\u308b\u3002<\/li>\n\n\n\n<li>\u5408\u6210\u30c7\u30fc\u30bf\u3092\u4ecb\u3057\u3066\u306e\u81ea\u5df1\u6539\u5584\u306f\u3084\u306f\u308a\u6709\u52b9\u305d\u3046\u3068\u3044\u3046\u5370\u8c61\u3002\uff08\u4eca\u3067\u3082\u4e00\u5b9a\u5b9f\u7528\u7684\u3067\u3042\u308b\u3068\u601d\u3046\u304c\uff09AGI\u3068\u304b\u3044\u3046\u4e16\u754c\u89b3\u3092\u8003\u3048\u308b\u3068\u6642\u9593\u7684\u5236\u7d04\u304c\u89e3\u6d88\u3067\u304d\u308b\u304b\u304c\u30dd\u30a4\u30f3\u30c8\u3060\u308d\u3046\u304b\u3002\uff08AI\u306b\u3082\u7761\u7720\u304c\u5fc5\u8981\u3068\u8a00\u3044\u3064\u3064\u3053\u306e\u624b\u306e\u51e6\u7406\u3092\u884c\u3046\u3088\u3046\u306a\u5c11\u3057\u672a\u6765\u304c\u5984\u60f3\u3055\u308c\u308b\uff09<\/li>\n\n\n\n<li>\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u30b5\u30a4\u30c8\u306f<a href=\"https:\/\/jyopari.github.io\/posts\/seal\">Self-Adapting Language Models<\/a><\/li>\n<\/ul>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Self-Adapting Improvement Loops for Robotic Learning\u00a0<\/strong>[30.8]<br>\u5c02\u9580\u5bb6\u306b\u3088\u308b\u30c7\u30e2\u30f3\u30b9\u30c8\u30ec\u30fc\u30b7\u30e7\u30f3\u3067\u8a13\u7df4\u3055\u308c\u305f\u30d3\u30c7\u30aa\u751f\u6210\u30e2\u30c7\u30eb\u306f\u3001\u30ed\u30dc\u30c3\u30c8\u30bf\u30b9\u30af\u3092\u89e3\u304f\u305f\u3081\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u306e\u9ad8\u3044\u30c6\u30ad\u30b9\u30c8\u6761\u4ef6\u4ed8\u304d\u30d3\u30b8\u30e5\u30a2\u30eb\u30d7\u30e9\u30f3\u30ca\u30fc\u3068\u3057\u3066\u5229\u7528\u3055\u308c\u3066\u304d\u305f\u3002 \u672c\u7814\u7a76\u3067\u306f,\u81ea\u5df1\u751f\u6210\u30c8\u30e9\u30b8\u30a7\u30af\u30c8\u30ea\u4e0a\u3067,\u30c9\u30e1\u30a4\u30f3\u5185\u30d3\u30c7\u30aa\u30e2\u30c7\u30eb\u3092\u53cd\u5fa9\u7684\u306b\u66f4\u65b0\u3059\u308b\u81ea\u5df1\u6539\u5584\u30eb\u30fc\u30d7(SAIL)\u3092\u63d0\u6848\u3059\u308b\u3002 \u5f93\u6765\u306e\u30c9\u30e1\u30a4\u30f3\u5185\u30d3\u30c7\u30aa\u30e2\u30c7\u30eb\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3067\u306f,\u65b0\u898f\u30bf\u30b9\u30af\u306e\u7e70\u308a\u8fd4\u3057\u306b\u5bfe\u3057\u3066,\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u304c\u7d99\u7d9a\u7684\u306b\u5411\u4e0a\u3059\u308b\u3053\u3068\u304c\u78ba\u8a8d\u3067\u304d\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2506.06658v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2506.06658v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Sat, 07 Jun 2025 04:34:37 GMT)<\/li>\n\n\n\n<li>\u300cwe highlight that adaptation with large-scale pretrained text-conditioned video models is critical for facilitating self-improvement, by contributing text-conditioned generalization capabilities and motion priors.\u300d\u3068\u3053\u3061\u3089\u306f\u52d5\u753b\u751f\u6210\u30e2\u30c7\u30eb\u3092\u6d3b\u7528\u3059\u308b\u30a2\u30d7\u30ed\u30fc\u30c1\u3002<\/li>\n\n\n\n<li>\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u30b5\u30a4\u30c8\u306f<a href=\"https:\/\/diffusion-supervision.github.io\/sail\/\">SAIL<\/a><\/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":[356,390,576],"class_list":["post-6931","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-self-x","tag-synthetic-data","tag-576"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6931","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=6931"}],"version-history":[{"count":2,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6931\/revisions"}],"predecessor-version":[{"id":6936,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6931\/revisions\/6936"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6931"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6931"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6931"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}