{"id":4638,"date":"2024-04-01T05:38:00","date_gmt":"2024-03-31T20:38:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=4638"},"modified":"2024-04-01T05:38:00","modified_gmt":"2024-03-31T20:38:00","slug":"safe-search-augmented-factuality-evaluator","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=4638","title":{"rendered":"SAFE: Search-Augmented Factuality Evaluator"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Long-form factuality in large language models\u00a0<\/strong>[59.3]<br>\u5927\u898f\u6a21\u8a00\u8a9e\u30e2\u30c7 \u30eb(LLM)\u306f\u3001\u3057\u3070\u3057\u3070\u3001\u30aa\u30fc\u30d7\u30f3\u30a8\u30f3\u30c9\u30c8\u30d4\u30c3\u30af\u306e\u4e8b\u5b9f\u691c\u7d22\u30d7\u30ed\u30f3\u30d7\u30c8\u306b\u5fdc\u7b54\u3059\u308b\u3068\u304d\u306b\u3001\u4e8b\u5b9f\u30a8\u30e9\u30fc\u3092\u542b\u3080\u30b3\u30f3\u30c6\u30f3\u30c4\u3092\u751f\u6210\u3059\u308b\u3002 \u307e\u305a\u6700\u521d\u306bGPT-4\u3092\u7528\u3044\u3066\u300138\u306e\u30c8\u30d4\u30c3\u30af\u306b\u307e\u305f\u304c\u308b\u4f55\u5343\u3082\u306e\u8cea\u554f\u304b\u3089\u306a\u308b\u30d7\u30ed\u30f3\u30d7\u30c8\u30bb\u30c3\u30c8\u3067\u3042\u308bLongFact\u3092\u751f\u6210\u3057\u307e\u3059\u3002 \u305d\u3053\u3067\u6211\u3005\u306f,LLM\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u3092\u691c\u7d22\u62e1\u5f35\u73fe\u5b9f\u6027\u8a55\u4fa1\u5668 (SAFE) \u3068\u547c\u3076\u624b\u6cd5\u306b\u3088\u308a,\u9577\u671f\u7684\u4e8b\u5b9f\u6027\u306e\u81ea\u52d5\u8a55\u4fa1\u5668\u3068\u3057\u3066\u4f7f\u7528\u3067\u304d\u308b\u3053\u3068\u3092\u63d0\u6848\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2403.18802v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2403.18802v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Wed, 27 Mar 2024 17:48:55 GMT)<\/li>\n\n\n\n<li>\u4e8b\u5b9f\u6027\u306e\u9593\u9055\u3044\u3092\u91cd\u8996\u3057\u305f\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3001\u300cSAFE utilizes an LLM to break down a long-form response into a set of individual facts and to evaluate the accuracy of each fact using a multi-step reasoning process comprising sending search queries to Google Search and determining whether a fact is supported by the search results.\u300d\u300cEmpirically, we demonstrated that SAFE achieves superhuman performance by agreeing with 72% of human annotations and winning 76% of examples out of a set of 100 randomly-sampled disagreement cases.\u300d\u3068\u306e\u3053\u3068\u3002\u30d9\u30f3\u30c1\u30de\u30fc\u30af\u3068\u3057\u3066\u306e\u8a55\u4fa1\u7d50\u679c\u306fGPT-4-turbo > Gemini Ultra > Calude-3 OPUS\u3067Claude 3 OPUS\u306f\u30cf\u30eb\u30b7\u30cd\u30fc\u30b7\u30e7\u30f3\u304c\u591a\u3044\u306e\u3067\u306f\uff1f\u3068\u3044\u3046\u5370\u8c61\u3092\u88cf\u4ed8\u3051\u3066\u3044\u305d\u3046\u306b\u601d\u3046\u3002SAFE\u306f\u8a55\u4fa1\u7528\u3060\u3051\u3067\u306a\u304f\u4e8c\u6b21\u30c1\u30a7\u30c3\u30af\u306b\u3082\u6709\u7528\u305d\u3046\u3002<\/li>\n\n\n\n<li>\u30ea\u30dd\u30b8\u30c8\u30ea\u306f<a href=\"https:\/\/github.com\/google-deepmind\/long-form-factuality\">google-deepmind\/long-form-factuality: Benchmarking long-form factuality in large language models. Original code for our paper &#8220;Long-form factuality in large language models.&#8221; (github.com)<\/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":[184,223,517],"class_list":["post-4638","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-hallucination","tag-llm","tag-517"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4638","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=4638"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4638\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4638"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4638"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4638"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}