{"id":8295,"date":"2026-03-04T03:52:00","date_gmt":"2026-03-03T18:52:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=8295"},"modified":"2026-02-28T16:07:20","modified_gmt":"2026-02-28T07:07:20","slug":"counterfactual-simulation-training-for-chain-of-thought-faithfulness","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=8295","title":{"rendered":"Counterfactual Simulation Training for Chain-of-Thought Faithfulness\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Counterfactual Simulation Training for Chain-of-Thought Faithfulness\u00a0<\/strong>[46.3]<br>\u6211\u3005\u306f,CST(Counterfactual Simulation Training)\u3068\u547c\u3070\u308c\u308b\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u624b\u6cd5\u3092\u5c0e\u5165\u3059\u308b\u3002 CST\u306f\u3001\u30b7\u30df\u30e5\u30ec\u30fc\u30bf\u30fc\u304c\u507d\u306e\u5165\u529b\u306b\u5bfe\u3057\u3066\u30e2\u30c7\u30eb\u306e\u51fa\u529b\u3092\u6b63\u78ba\u306b\u4e88\u6e2c\u3067\u304d\u308bCoT\u306b\u5831\u916c\u3092\u4e0e\u3048\u308b\u3002 \u6700\u5927235B\u30d1\u30e9\u30e1\u30fc\u30bf\u306e\u30e2\u30c7\u30eb\u306b\u3088\u308b\u5b9f\u9a13\u306b\u3088\u308a\u3001CST\u306f\u30ad\u30e5\u30fc\u30d9\u30fc\u30b9\u306e\u30ab\u30a6\u30f3\u30bf\u30fc\u30d5\u30a1\u30af\u30c8\u306e\u7cbe\u5ea6\u3092\u5927\u5e45\u306b\u5411\u4e0a\u3067\u304d\u308b\u3053\u3068\u304c\u793a\u3055\u308c\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2602.20710v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2602.20710v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Tue, 24 Feb 2026 09:15:30 GMT)<\/li>\n\n\n\n<li>CoT\u306e\u4fe1\u983c\u6027\u3092\u5411\u4e0a\u3055\u305b\u308b\u305f\u3081\u300cwe introduce a training method called Counterfactual Simulation Training (CST), which aims to improve CoT faithfulness by rewarding CoTs that enable a simulator to accurately predict a model\u2019s outputs over counterfactual inputs. We apply CST in two settings: (1) CoT monitoring with cue-based counterfactuals, to detect when models rely on spurious features, reward hack, or are sycophantic, and (2) counterfactual simulation over generic model-based counterfactuals, to encourage models to produce more faithful, generalizable reasoning in the CoT.\u300d\u3068\u3044\u3046\u30a2\u30d7\u30ed\u30fc\u30c1\u3092\u63d0\u6848\u3002Reasoning\u306e\u904e\u7a0b\u3092\u30b3\u30f3\u30c8\u30ed\u30fc\u30eb\u3059\u308b\u306e\u3082\u91cd\u8981\u306a\u306e\u306f\u305d\u3046\u3060\u3068\u601d\u3046\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/peterbhase\/counterfactual-simulation-training\">GitHub &#8211; peterbhase\/counterfactual-simulation-training: Codebase for paper: &#8220;Counterfactual Simulation Training for Chain-of-Thought Faithfulness&#8221;<\/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":[59],"class_list":["post-8295","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-chain-of-thought"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8295","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=8295"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8295\/revisions"}],"predecessor-version":[{"id":8296,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/8295\/revisions\/8296"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8295"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8295"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8295"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}