{"id":4687,"date":"2024-04-08T05:43:00","date_gmt":"2024-04-07T20:43:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=4687"},"modified":"2024-04-08T05:43:00","modified_gmt":"2024-04-07T20:43:00","slug":"against-the-achilles-heel-a-survey-on-red-teaming-for-generative-models","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=4687","title":{"rendered":"Against The Achilles&#8217; Heel: A Survey on Red Teaming for Generative Models\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Against The Achilles&#8217; Heel: A Survey on Red Teaming for Generative Models&nbsp;<\/strong>[60.2]<br>\u8d64\u3044\u30c1\u30fc\u30e0\u4f5c\u308a\u306e\u5206\u91ce\u306f\u6025\u901f\u306b\u6210\u9577\u3057\u3066\u304a\u308a\u3001\u30d1\u30a4\u30d7\u30e9\u30a4\u30f3\u5168\u4f53\u3092\u30ab\u30d0\u30fc\u3059\u308b\u5305\u62ec\u7684\u306a\u7d44\u7e54\u306e\u5fc5\u8981\u6027\u3092\u5f37\u8abf\u3057\u3066\u3044\u308b\u3002 120\u4ee5\u4e0a\u306e\u8ad6\u6587\u3092\u8abf\u67fb\u3057,\u8a00\u8a9e\u30e2\u30c7\u30eb\u56fa\u6709\u306e\u80fd\u529b\u306b\u6839\u3056\u3057\u305f,\u304d\u3081\u7d30\u304b\u3044\u653b\u6483\u6226\u7565\u306e\u5206\u985e\u3092\u5c0e\u5165\u3057\u305f\u3002 \u6211\u3005\u306f,\u69d8\u3005\u306a\u81ea\u52d5\u30ec\u30c3\u30c9\u30fb\u30c1\u30fc\u30e0\u30fb\u30a2\u30d7\u30ed\u30fc\u30c1\u3092\u7d71\u5408\u3059\u308b\u30b5\u30fc\u30c1\u30fb\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3092\u958b\u767a\u3057\u305f\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2404.00629v1\">\u8ad6\u6587<\/a>&nbsp;&nbsp;<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2404.00629v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>&nbsp; &nbsp;(Sun, 31 Mar 2024 09:50:39 GMT)<\/li>\n\n\n\n<li>\u793e\u4f1a\u5b9f\u88c5\u306b\u304a\u3044\u3066\u91cd\u8981\u306aRed Teaming\u306b\u95a2\u3059\u308b\u30b5\u30fc\u30d9\u30a4\u3002\u300cFigure 2: An overview of GenAI red teaming flow.\u300d\u304b\u3089\u59cb\u307e\u308b\u69cb\u6210\u304c\u308f\u304b\u308a\u3084\u3059\u3044\u3002CC-BY\u306a\u306e\u3082\u3046\u308c\u3057\u3044\u3068\u3053\u308d\u3002<\/li>\n<\/ul>\n\n\n\n<p><a href=\"https:\/\/www.anthropic.com\/research\/many-shot-jailbreaking\">Many-shot jailbreaking \\ Anthropic<\/a><br>\u6211\u3005\u306f\u3001\u6709\u7528\u3067\u7121\u5bb3\u3067\u6b63\u76f4\u306aAI\u30a2\u30b7\u30b9\u30bf\u30f3\u30c8\u3092\u30bf\u30fc\u30b2\u30c3\u30c8\u306b\u3057\u305f\u591a\u767a\u30b8\u30a7\u30a4\u30eb\u30d6\u30ec\u30fc\u30ad\u30f3\u30b0(MSJ)\u3092\u7814\u7a76\u3057\u305f\u3002MSJ\u306f\u6570\u767a\u306e\u30b8\u30a7\u30a4\u30eb\u30d6\u30ec\u30fc\u30af\u306e\u6982\u5ff5\u3092\u62e1\u5f35\u3057\u3001\u653b\u6483\u8005\u306f\u30e2\u30c7\u30eb\u304c\u901a\u5e38\u7b54\u3048\u308b\u3053\u3068\u3092\u62d2\u5426\u3059\u308b\u4e00\u9023\u306e\u30af\u30a8\u30ea\u3092\u542b\u3080\u67b6\u7a7a\u306e\u5bfe\u8a71\u3067\u30e2\u30c7\u30eb\u3092\u30d7\u30ed\u30f3\u30d7\u30c8\u3059\u308b\u3002<br>\u300cWe found that the effectiveness of attacks, and of in-context learning more generally, could be characterized by simple power laws.\u300d\u3068\u3044\u3046\u3068\u3066\u3082\u30b7\u30f3\u30d7\u30eb\u306a\u653b\u6483\u304c\u6709\u52b9\u3067\u3042\u3063\u305f\u308a\u3082\u3057\u3066\u653b\u6483\u6226\u7565\u3082\u65e5\u3005\u9032\u5316\u3057\u3066\u3044\u308b\u72b6\u6cc1\u3067\u5b89\u5168\u6027\u3092\u78ba\u4fdd\u3057\u3066\u3044\u304f\u306e\u306f\u3068\u3066\u3082\u5927\u5909\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Many-shot jailbreaking \\ Anthropic\u6211\u3005\u306f\u3001\u6709\u7528\u3067\u7121\u5bb3\u3067\u6b63\u76f4\u306aAI\u30a2\u30b7\u30b9\u30bf\u30f3\u30c8\u3092\u30bf\u30fc\u30b2\u30c3\u30c8\u306b\u3057\u305f\u591a\u767a\u30b8\u30a7\u30a4\u30eb\u30d6\u30ec\u30fc\u30ad\u30f3\u30b0(MSJ)\u3092\u7814\u7a76\u3057\u305f\u3002MSJ\u306f\u6570\u767a\u306e\u30b8\u30a7\u30a4\u30eb\u30d6\u30ec\u30fc\u30af\u306e\u6982\u5ff5\u3092\u62e1\u5f35\u3057\u3001 &hellip; <a href=\"https:\/\/devneko.jp\/wordpress\/?p=4687\" class=\"more-link\"><span class=\"screen-reader-text\">&#8220;Against The Achilles&#8217; Heel: A Survey on Red Teaming for Generative Models\u00a0&#8221; \u306e<\/span>\u7d9a\u304d\u3092\u8aad\u3080<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[32,387,622],"class_list":["post-4687","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-attack","tag-survey","tag-622"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4687","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=4687"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4687\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4687"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4687"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4687"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}