{"id":8038,"date":"2026-01-05T03:52:00","date_gmt":"2026-01-04T18:52:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=8038"},"modified":"2026-01-04T20:02:24","modified_gmt":"2026-01-04T11:02:24","slug":"training-ai-co-scientists-using-rubric-rewards","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=8038","title":{"rendered":"Training AI Co-Scientists Using Rubric Rewards"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Training AI Co-Scientists Using Rubric Rewards\u00a0<\/strong>[36.9]<br>AI\u306e\u5171\u540c\u7814\u7a76\u8005\u306e\u91cd\u8981\u306a\u7279\u5fb4\u306f\u3001\u76ee\u7684\u3068\u5236\u7d04\u306e\u30bb\u30c3\u30c8\u304b\u3089\u7814\u7a76\u8a08\u753b\u3092\u751f\u6210\u3059\u308b\u80fd\u529b\u3067\u3042\u308b\u3002 \u672c\u7814\u7a76\u3067\u306f,\u65e2\u5b58\u306e\u7814\u7a76\u8ad6\u6587\u306e\u81a8\u5927\u306a\u30b3\u30fc\u30d1\u30b9\u3092\u6d3b\u7528\u3057\u3066,\u3088\u308a\u826f\u3044\u7814\u7a76\u8a08\u753b\u3092\u751f\u307f\u51fa\u3059\u8a00\u8a9e\u30e2\u30c7\u30eb\u3092\u8a13\u7df4\u3059\u308b\u65b9\u6cd5\u306b\u3064\u3044\u3066\u691c\u8a0e\u3059\u308b\u3002 \u8907\u6570\u306e\u30c9\u30e1\u30a4\u30f3\u306b\u308f\u305f\u308b\u8ad6\u6587\u304b\u3089\u7814\u7a76\u76ee\u6a19\u3068\u76ee\u6a19\u56fa\u6709\u306e\u30b0\u30eb\u30fc\u30d6\u3092\u81ea\u52d5\u62bd\u51fa\u3059\u308b\u3053\u3068\u3067\u3001\u30b9\u30b1\u30fc\u30e9\u30d6\u30eb\u3067\u591a\u69d8\u306a\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u30b3\u30fc\u30d1\u30b9\u3092\u69cb\u7bc9\u3057\u307e\u3059\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2512.23707v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2512.23707v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 29 Dec 2025 18:59:33 GMT)<\/li>\n\n\n\n<li>\u300cwe leverage existing scientific papers to improve language models at generating research plans for diverse open-ended research goals. We propose a scalable training procedure that uses a language model to extract research goals and grading rubrics from papers, and trains the plan generator with self-grading using the goal-specific rubrics as privileged information.\u300d\u3068\u306e\u3053\u3068\u3067\u65e2\u5b58\u306e\u7814\u7a76\u8ad6\u6587\u3092\u7528\u3044\u3066LRM\u306e\u7814\u7a76\u8a08\u753b\u518d\u751f\u80fd\u529b\u3092\u5f37\u5316\u3002\u00a0Qwen-3-30B-A3B-Instruct\u30d9\u30fc\u30b9\u3067\u3042\u308b\u3053\u3068\u3092\u8003\u3048\u308b\u3068\u300cThe obtained performance makes our 30B model competitive with Grok-4-Thinking (xAI, 2025), though it remains behind the best performing model, GPT-5-Thinking (OpenAI, 2025).\u300d\u306f\u5065\u95d8\u3057\u3066\u3044\u308b\u3088\u3046\u306b\u601d\u3048\u308b\u3002<\/li>\n\n\n\n<li>\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u304c\u516c\u958b\u3055\u308c\u3066\u3044\u308b\u3000<a href=\"https:\/\/huggingface.co\/datasets\/facebook\/research-plan-gen\">facebook\/research-plan-gen \u00b7 Datasets at Hugging Face<\/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":[623],"class_list":["post-8038","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-623"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8038","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=8038"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8038\/revisions"}],"predecessor-version":[{"id":8039,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8038\/revisions\/8039"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8038"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8038"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8038"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}