{"id":4308,"date":"2024-01-15T06:16:00","date_gmt":"2024-01-14T21:16:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=4308"},"modified":"2024-01-15T06:16:00","modified_gmt":"2024-01-14T21:16:00","slug":"language-models-understand-numbers-at-least-partially","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=4308","title":{"rendered":"Language Models Understand Numbers, at Least Partially"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>Language Models Understand Numbers, at Least Partially\u00a0<\/strong>[32.5]<br>\u6570\u5b66\u7684\u554f\u984c\u306b\u304a\u3051\u308b\u57fa\u672c\u8981\u7d20\u3068\u306a\u308b\u6570\u306b\u3064\u3044\u3066,\u8a00\u8a9e\u30e2\u30c7\u30eb\u304c\u7406\u89e3\u3057\u3066\u3044\u308b\u304b\u3069\u3046\u304b\u3092\u8003\u5bdf\u3059\u308b\u3002 \u672c\u7814\u7a76\u3067\u306f,\u4ed8\u52a0\u554f\u984c\u3092\u542b\u3080\u5408\u6210\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u3092\u69cb\u7bc9\u3057,\u7dda\u5f62\u30d7\u30ed\u30fc\u30d6\u3092\u7528\u3044\u3066\u30e2\u30c7\u30eb\u306e\u96a0\u308c\u72b6\u614b\u304b\u3089\u5165\u529b\u756a\u53f7\u3092\u8aad\u307f\u53d6\u308b\u3002 \u4e88\u5099\u7814\u7a76\u306f\u3001\u8a00\u8a9e\u30e2\u30c7\u30eb\u304c\u6570\u306e\u90e8\u5206\u7684\u7406\u89e3\u3092\u793a\u3059\u3053\u3068\u3092\u793a\u5506\u3057\u3066\u3044\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2401.03735v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2401.03735v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Mon, 8 Jan 2024 08:54:22 GMT)<\/li>\n\n\n\n<li>LLM\u304c\u6570\u5024\u3092\u7406\u89e3\u3057\u3066\u3044\u308b\u304b\uff1f\u306e\u5206\u6790\u3002\u5229\u7528\u3057\u3066\u3044\u308b\u7acb\u5834\u3060\u3068\u602a\u3057\u3044\u3068\u601d\u3063\u3066\u3044\u305f\u304c\u3001\u300cExperimental results prove that LLMs do have a rough estimation of input numbers in their hidden states, but the compression process may not be lossless.\u300d\u3068\u5b8c\u5168\u3067\u306f\u306a\u3044\u304c\u80af\u5b9a\u7684\u306a\u7d50\u679c\u306e\u3088\u3046\u3002\u300cLLMs exhibit the ability to utilize compressed numbers to perform arithmetic calculations, and the ability to perform calculations is explicitly related to the scale of models.\u300d\u3068\u3044\u3046\u306e\u3082\u3068\u3066\u3082\u8208\u5473\u6df1\u3044\u3002<\/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":[223],"class_list":["post-4308","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-llm"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4308","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=4308"}],"version-history":[{"count":0,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4308\/revisions"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4308"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4308"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4308"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}