{"id":6887,"date":"2025-06-09T05:13:00","date_gmt":"2025-06-08T20:13:00","guid":{"rendered":"https:\/\/devneko.jp\/wordpress\/?p=6887"},"modified":"2025-06-07T21:17:29","modified_gmt":"2025-06-07T12:17:29","slug":"how-much-do-language-models-memorize","status":"publish","type":"post","link":"https:\/\/devneko.jp\/wordpress\/?p=6887","title":{"rendered":"How much do language models memorize?\u00a0"},"content":{"rendered":"\n<ul class=\"wp-block-list\">\n<li><strong>How much do language models memorize?\u00a0<\/strong>[104.2]<br>\u6211\u3005\u306f\u8a18\u61b6\u30922\u3064\u306e\u69cb\u6210\u8981\u7d20\u306b\u5206\u3051\u3066\u3044\u308b:\u300c\u6587\u4f53\u8a18\u61b6\u300d\u3068\u300c\u6587\u4f53\u4e00\u822c\u5316\u300d\u3067\u3042\u308b\u3002 \u4e00\u822c\u5316\u3092\u5b8c\u5168\u306b\u6392\u9664\u3059\u308b\u3068\u3001\u30e2\u30c7\u30eb\u30ad\u30e3\u30d1\u30b7\u30c6\u30a3\u3092\u898b\u7a4d\u3082\u308b\u30c8\u30fc\u30bf\u30eb\u30fb\u30e1\u30e2\u30ea\u5316\u3092\u8a08\u7b97\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u308b\u3002 \u30b5\u30a4\u30ba\u304c\u5927\u304d\u304f\u306a\u308b\u30c7\u30fc\u30bf\u30bb\u30c3\u30c8\u4e0a\u3067\u8a00\u8a9e\u30e2\u30c7\u30eb\u3092\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3057\u3001\u30ad\u30e3\u30d1\u30b7\u30c6\u30a3\u304c\u6e80\u305f\u3055\u308c\u308b\u307e\u3067\u30e2\u30c7\u30eb\u3092\u8a18\u61b6\u3057\u3001\u305d\u306e\u6642\u70b9\u3067\u306e\u300c\u30b0\u30eb\u30fc\u30ad\u30f3\u30b0\u300d\u304c\u59cb\u307e\u308a\u3001\u30e2\u30c7\u30eb\u304c\u4e00\u822c\u5316\u3057\u59cb\u3081\u308b\u306b\u3064\u308c\u3066\u610f\u56f3\u3057\u306a\u3044\u8a18\u61b6\u304c\u6e1b\u5c11\u3059\u308b\u306e\u3092\u89b3\u5bdf\u3059\u308b\u3002<br><a href=\"http:\/\/arxiv.org\/abs\/2505.24832v1\">\u8ad6\u6587<\/a>\u00a0\u00a0<a href=\"https:\/\/fugumt.com\/fugumt\/paper_check\/2505.24832v1\">\u53c2\u8003\u8a33\uff08\u30e1\u30bf\u30c7\u30fc\u30bf\uff09<\/a>\u00a0 \u00a0(Fri, 30 May 2025 17:34:03 GMT)<\/li>\n\n\n\n<li>AGI\u3092\u76ee\u6307\u3059\u3046\u3048\u3067\u3068\u3066\u3082\u91cd\u8981\u306a\u8a18\u61b6\u306b\u95a2\u3059\u308b\u5831\u544a\u3001\u300cWe formally separate memorization into two components: unintended memorization, the information a model contains about a specific dataset, and generalization, the information a model contains about the true data-generation process. When we completely eliminate generalization, we can compute the total memorization, which provides an estimate of model capacity: our measurements estimate that GPT-style models have a capacity of approximately 3.6 bits per parameter.\u300d\u3068\u306e\u3053\u3068\u3002<\/li>\n\n\n\n<li><a href=\"https:\/\/devneko.jp\/wordpress\/?p=4776\">\u5f15\u7528\u3055\u308c\u3066\u3044\u308b\u304c\u3001Physics of Language Models: Part 3.3, Knowledge Capacity Scaling Laws\u00a0 \u2013 arXiv\u6700\u65b0\u8ad6\u6587\u306e\u7d39\u4ecb<\/a>\u306a\u3069\u3001\u3053\u306e\u624b\u306e\u7814\u7a76\u306f\u672c\u5f53\u306b\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":[244],"class_list":["post-6887","post","type-post","status-publish","format-standard","hentry","category-arxiv","tag-memory"],"_links":{"self":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6887","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=6887"}],"version-history":[{"count":1,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6887\/revisions"}],"predecessor-version":[{"id":6888,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/6887\/revisions\/6888"}],"wp:attachment":[{"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6887"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6887"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/devneko.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6887"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}