{"id":5685,"date":"2024-10-30T05:10:37","date_gmt":"2024-10-29T20:10:37","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=5685"},"modified":"2024-10-30T05:10:37","modified_gmt":"2024-10-29T20:10:37","slug":"hallueditbench%e3%80%81should-we-really-edit-language-models-on-the-evaluation-of-edited-language-models","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=5685","title":{"rendered":"HalluEditBench\u3001Should We Really Edit Language Models? On the Evaluation of Edited Language Models\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Should We Really Edit Language Models? On the Evaluation of Edited Language Models\u00a0<\/strong>[15.6]<br>\u65e2\u5b58\u306e\u7de8\u96c6\u624b\u6cd5\u306f\u3001\u4e00\u822c\u7684\u306a\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3067\u5fc5\u7136\u7684\u306b\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u304c\u4f4e\u4e0b\u3059\u308b\u3002 \u30a4\u30f3\u30b9\u30c8\u30e9\u30af\u30b7\u30e7\u30f3\u30c1\u30e5\u30fc\u30cb\u30f3\u30b0\u3055\u308c\u305f\u30e2\u30c7\u30eb\u306f\u3001\u7de8\u96c6\u304c\u3088\u308a\u5805\u7262\u3067\u3001\u7de8\u96c6\u5f8c\u306e\u4e00\u822c\u7684\u306a\u77e5\u8b58\u306b\u5bfe\u3059\u308b\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u4f4e\u4e0b\u304c\u5c11\u306a\u3044\u3002 \u305d\u306e\u7d50\u679c,\u73fe\u5728\u306e\u7de8\u96c6\u624b\u6cd5\u306f,\u8a00\u8a9e\u30e2\u30c7\u30eb\u5185\u306e\u5c0f\u898f\u6a21\u306a\u77e5\u8b58\u66f4\u65b0\u306b\u306e\u307f\u9069\u3057\u3066\u3044\u308b\u3053\u3068\u304c\u308f\u304b\u3063\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2410.18785v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2410.18785v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Thu, 24 Oct 2024 14:36:48 GMT)<\/li>\n\n\n\n<li>\u77e5\u8b58\u7de8\u96c6\u306b\u95a2\u3059\u308b\u5206\u6790\u3001\u300cThe experimental results indicate that existing editing methods can preserve the general capabilities of the model within a limited number of edits, not exceeding a few dozen.\u300d\u3001\u300cOur experiments demonstrate that after only a few dozen edits, the safety of the model is compromised, including those models that have been aligned.\u300d\u3068\u3044\u3046\u6307\u6458\u3002\u307e\u305f\u3001\u300cLanguage model with large scale is more resistant to editing compared to small model.\u300d\u3068\u3044\u3046\u306e\u3082\u3001Knowledge Editing\u306e\u7814\u7a76\u6210\u679c\u304c\u5b9f\u554f\u984c\u306b\u9069\u7528\u56f0\u96e3\u3067\u3042\u308b\u53ef\u80fd\u6027\u3092\u793a\u5506\u3057\u3066\u3044\u308b\u3068\u601d\u3046\u3002<\/li>\n<\/ul>\n\n\n\n<p><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Can Knowledge Editing Really Correct Hallucinations?\u00a0<\/strong>[16.3]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7\u30eb(LLM)\u306f\u3001\u30bf\u30b9\u30af\u306b\u307e\u305f\u304c\u308b\u512a\u308c\u305f\u80fd\u529b\u306b\u3082\u304b\u304b\u308f\u3089\u305a\u3001\u751f\u6210\u3055\u308c\u305f\u30b3\u30f3\u30c6\u30f3\u30c4\u306e\u975e\u73fe\u5b9f\u7684\u306a\u60c5\u5831\u3092\u53c2\u7167\u3057\u306a\u304c\u3089\u5e7b\u899a\u306b\u60a9\u307e\u3055\u308c\u308b\u3002 \u77e5\u8b58\u7de8\u96c6\u306f,LLM\u3067\u7b26\u53f7\u5316\u3055\u308c\u305f\u8aa4\u3063\u305f\u4e8b\u5b9f\u77e5\u8b58\u3092,\u30b9\u30af\u30e9\u30c3\u30c1\u304b\u3089\u30ea\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3092\u907f\u3051\u308b\u3068\u3044\u3046\u5229\u70b9\u306b\u3088\u3063\u3066\u88dc\u6b63\u3059\u308b,\u65b0\u3057\u3044\u4e00\u822c\u7684\u306a\u30d1\u30e9\u30c0\u30a4\u30e0\u3068\u3057\u3066\u958b\u767a\u3055\u308c\u305f\u3002 \u5b9f\u4e16\u754c\u306e\u5e7b\u899a\u3092\u6b63\u3059\u305f\u3081\u306e\u77e5\u8b58\u7de8\u96c6\u624b\u6cd5\u3092\u5168\u4f53\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3059\u308b\u305f\u3081\u306b,HaluEditBench\u3092\u63d0\u6848\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2410.16251v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2410.16251v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 21 Oct 2024 17:55:54 GMT)<\/li>\n\n\n\n<li>Knowledge Editing\u306b\u95a2\u3059\u308b\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u306e\u63d0\u6848\u3001\u300cThe effectiveness of knowledge editing methods in correcting real-world hallucinations could be far from what their performance on existing datasets suggests, reflecting the potential unreliability of current assessment of different knowledge editing techniques.\u300d\u3068\u6c17\u306b\u306a\u308b\u6307\u6458\u304c\u3042\u308b\u3002<\/li>\n\n\n\n<li><a href=\"https:\/\/llm-editing.github.io\/\">\u30ea\u30dd\u30b8\u30c8\u30ea\u306fEditing LLMs<\/a><\/li>\n<\/ul>\n\n\n\n<p><\/p>\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":[213],"class_list":["post-5685","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-knowledge-editing"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5685","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=5685"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/5685\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=5685"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=5685"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=5685"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}